# Applied AI for Biotech R&D and Biomanufacturing — Upskilling Plan v4

**Author:** Peter Ciancaglini · **Date:** 2026-09-11 · **Horizon:** 20 weeks at 8-10 h/week, then a 6-month commercial phase
**Replaces:** v3 (Vara, 2026-09-10). Vara is not started as a flagship; two of its ideas (telemetry schema, calibrated judge) are reused here.
**Working name of the program:** `labloop` (rename freely; the name is not the point, the number is)

---

## 0. Read this first (what changed and why)

| # | Topic | v3 (Vara) | v4 (this plan) | Origin |
|---|---|---|---|---|
| 1 | Audience | Devs, NGOs, Spanish-speaking startups in general | Wet-lab biotech, biomanufacturing, agbio and pharma R&D: organizations with money, complex problems and a slow, expensive experimental loop | Your ask + the GridX list (64 of 86 startups have no AI/IoT base) |
| 2 | Flagship | One benchmark (Vara) | Three small **instruments** that each close one gap the list shows repeatedly, plus one priced offer | Your synthesis: instruments are your talent, selling them is your trap |
| 3 | Theory | LLM internals (Karpathy, CS336) | Bioprocess, design of experiments, Bayesian optimization, multivariate monitoring, hybrid models, document AI on-prem, plus a science literacy thread (proteins, molecules, single-cell, microbiome) | "Professionalize in AI science" |
| 4 | Money | Offer expected in month 6-9 | **First invoice by week 16** (any amount). The charging muscle is trained inside the plan, not after it | Synthesis: 25 years, USD 0 charged |
| 5 | Publishing | 1 build log/week, 3 deep dives | GitHub 3-4 commits/week, LinkedIn 2/week (Sunday build log + Wednesday gap note), Medium 4, arXiv/bioRxiv 2, Zenodo DOIs 3, 60-s videos 3 | Your ask |
| 6 | Hardware | Cloud APIs | RTX 3050 + Debian boxes are a **feature**: "your lab's AI runs on a USD 300 computer and your data never leaves the building" is the selling line for IP-sensitive labs | Your constraint, reframed |
| 7 | Culture goal | Not addressed | Explicit filter for the next employer/client: scientist-led, biology-paced, no delivery-pressure culture | EY exit |

Hours: 8-10 h/week as in v3. If you can do more, the calendar compresses; it never expands in scope.

---

## 1. Thesis

> **Most biotech startups do not have an AI problem. They have a data-to-decision problem.**
> Their experimental loop (design → run → measure → analyze → decide → next run) takes weeks because the data lives in spreadsheets, instrument exports, paper batch records and PDFs. The AI that matters to them is not a chatbot. It is the layer that closes that loop: structured capture, models that work on small and expensive data, and a decision every week instead of every quarter.
> **I build that layer, measure how many days it takes out of the loop, and publish the number.**

Why this is your niche and not someone else's:

1. **It is the same talent you showed at EY.** You made delivery visible and measurable when nobody asked. The lab equivalent is "how many days from question to decision, and where do those days go". Nobody in a wet lab has that number.
2. **The evidence is in the list.** 64 of 86 GridX startups have no AI/IoT base. Their stated gaps cluster into four repeated patterns (section 3), and none of the four needs a new algorithm. They need an engineer who understands the science, the software and the money.
3. **The money is real and it is not price-sensitive.** Pharma, CDMOs, agbio groups and funded biotech pay for speed (startups: runway) and for risk reduction (pharma: lost batches, audit findings). What your GridX contact told you ("if they are not burning, they are not growing") means: sell **weeks saved per experiment cycle**, converted into runway, not "cost savings". Startups buy time. Pharma buys risk.
4. **Biology sets the pace, not a partner.** A fermentation takes 7 days. A field trial takes a season. A stability study takes months. That timescale structurally prevents the "deliver more with less by Friday" culture you are leaving. Scientist-led companies are, on average, the calmest place a senior engineer can work.
5. **The AI-native biotechs are your peers, not your clients.** Tesabio.ai, Zavia, Nanogrow, CellCo, Dogma, Mavios and the other 16 "AI: yes" companies show what the sophisticated end of the market looks like. You learn from them, cite them, ask them to review or co-author, and sell to the 64 that are ten years behind them in data maturity.

The metric that defines the brand: **loop cycle time** (days from hypothesis to decision) and **data readiness** (% of experimental data that is machine-readable within 24 h of being generated). Every instrument you build moves one of the two. Every offer you sell measures one of the two first.

---

## 2. Strategy: how the dots connect

```mermaid
flowchart LR
  T[Theory: bioprocess, DoE, BO, MSPC, hybrid models, doc AI] --> I1
  T --> I2
  T --> I3
  I1[Instrument 1: batchlens\nfermentation batch intelligence] --> E[Evidence: repos, DOIs, 2 preprints, 4 deep dives, 3 videos]
  I2[Instrument 2: trialkit\ntrials and assays to decision] --> E
  I3[Instrument 3: labdocs\non-prem lab document intelligence] --> E
  V[Vara ideas: telemetry schema + calibrated judge] --> I1
  V --> I3
  E --> A1[Audience 1: 64 non-AI GridX startups\nfeedback, data, co-authors, first paid audits]
  E --> A2[Audience 2: established pharma, agbio, CDMOs, funds\nhigh-ticket audits and pilots]
  E --> A3[Audience 3: scientific-data companies\nemployee track, remote, relaxed]
  A1 --> O[Offer: Loop Audit, 2 weeks, fixed price]
  A2 --> O
  O --> P[Pilot 6-8 weeks] --> R[Retainer: fractional AI lead for R&D]
  A3 --> J[Job offers as fallback or by choice]
```

The chain in words:

- **Theory feeds instruments.** Every lecture in section 6 exists because a practice in section 7 needs it. Nothing is studied "for later".
- **Instruments produce evidence, not products.** Each instrument is used on public data first, then on one founder's sample data for free, then inside a paid audit. You never sell the instrument. You sell the number it produces about *their* loop.
- **Evidence opens three audiences at once with the same artifacts.** The startup list is your proving ground and your first invoices. Established companies and funds are the high ticket. Scientific-data companies (TetraScience, Uncountable, DataHow, Benchling and the like) are the employee track if consulting is slower than your patience.
- **The offer is fixed-price and small on purpose.** A 2-week Loop Audit is below most purchase thresholds and can be paid by card or a single invoice. It exists to train the muscle you have never used, and to generate the anonymized case study that sells the next one.
- **Funds are the channel.** GridX, SF500, CITES and similar funds each sit on 30-100 portfolio companies with the same four gaps. One "portfolio data-readiness package" sold to a fund puts you in front of ten startups without ten sales conversations. This is the "facilitator channel" from your August synthesis, applied to biotech.
- **The Vara work is not lost.** Its telemetry schema (cost × model × task) and its calibrated LLM-judge method are exactly what makes instruments 1 and 3 credible: measured, not claimed.

---

## 3. Target audience and companies

### 3.1 The four gap archetypes in the GridX list

I re-read all 86 descriptions and gaps and grouped them. Counts are companies whose *primary* gap falls in the archetype (full mapping in Appendix A).

| Code | Archetype | Typical stated gap | Companies (primary fit) | Instrument | Sells as |
|---|---|---|---|---|---|
| **A** | Process and telemetry: bioreactors, photobioreactors, digesters, incubation rooms, extrusion lines, cold chain | Manual monitoring, paper or spreadsheet batch records, no golden batch, no early deviation detection, no yield optimization | Michroma, Levya, Fungilife, NanoInGreen, POLYMERA, Protiva, Food4You, Bioplastix, BreakPET, Hiamet, Caligenia, ByBug, RADIAL, algaebio+, Algalife, Tomorrow, Einsted, CyanoMin, Alkemio, Enzyva, BioBlends, Nat4Bio, BioPlaster, Neocell, Calfix (weak) | **1 · batchlens** | Batch intelligence audit; soft sensor pilot; "next run" optimization |
| **B** | Field trials and validation in the field: agro bioinputs, aquaculture | Trials scattered in spreadsheets, subjective visual scoring, no consolidated efficacy report for growers and investors | Puna Bio, Nunatak, Bsafe, Semion, MicroIN, Qumir Nano, AQUIT, FeedVax | **2 · trialkit** (field module) | Trial data consolidation + statistically defensible efficacy report |
| **C** | Lab assays and preclinical R&D data: microplates, kinetics, dose-response, formulation stability | Excel-based analysis, slow reports, no correlation between process variables and outcome, fragmented lead-selection data | Resistia, APEXzymes, Antarka, Syocin, Arakion, Scitherm, BioClé, Lipock, Kresko, Nanotransfer, Invitrall, Bioeutectics, Vexxel, Werken Vac | **2 · trialkit** (plate module) | Assay pipeline + automated report; formulation data model |
| **D** | Traceability, QMS and regulated manufacturing: audit trails, batch genealogy, clean-room records, personalized lots | No digital audit trail, manual protocols, regulatory traceability demanded by agencies and absent | Eywa Biotech, AVaTAR MedTech, NanoproX, Apasomics, Synthetic Ocular Restoration, SphereBio, New Organs Biotech | **3 · labdocs** | On-prem document intelligence + ALCOA+ audit trail; deviation report drafting |
| **E** | Diagnostics and clinical reporting: point-of-care readers, microbiome and genetic reports, fertility clinics | Manual report generation, no portal, no integration with HIS/LIS, data security | CASPR, Gisens, Ubique, ZEV, NanoJump, Enteria, Microgenesis, PreGen, Embryoxite, Fecundis, GlycoX | **4 (phase 2, after week 20)** | Clinical report automation with validation; regulated, higher friction |
| **F** | AI-native and computational platforms | Software engineering and MLOps maturity, not AI | Tesabio.ai, Nanogrow, CellCo, Dogma, CellRep, Elytron, Zavia, Bitgenia, MIGMA, Naturannova, Mavios, Geoprot, Monte Caldera, Qnity, Beeflow, Biodiversity Intelligence, ArgenTag, OncoPrecision, Stämm, Sylvarum | none | Do not sell to. Learn from, cite, ask for review, co-author, endorse on arXiv |

Off-thesis: Cryosmetics (e-commerce recommender). Ignore.

Rules for using this list in public:

- **Never name a company's weakness in public.** The CSV's "falencias" column is your private hypothesis, not a fact. In posts, talk about *archetypes* ("a fungal pigment fermentation startup") and name companies only when praising them or after they engaged.
- **Tier the 64 non-AI companies by what you need from them**, not by what they can pay: (1) five feedback partners who will look at an instrument for 30 minutes, (2) two data partners who will share a sample dataset under NDA for a free run, (3) one co-author or arXiv endorser (most founders are CONICET scientists with publication records).

### 3.2 Beyond the list: who actually pays high ticket

I have **not** verified the current status, funding or openings of these companies. Treat this as a research list; each name needs 10 minutes of verification before outreach.

**Established Argentine and regional companies (money, regulated processes, slow data):**

| Company | Why | What to offer |
|---|---|---|
| Bioceres / Rizobacter (Rosario) | Agbio group with bioinputs, field trials at scale, own VC (SF500) | Trial data platform audit (B); portfolio package via SF500 |
| Insud Pharma: mAbxience, Elea, Chemo | Biosimilars manufacturing, GMP, PAT and data-integrity budgets | Batch intelligence (A) and audit-trail (D) pilots |
| Biogénesis Bagó, Sinergium Biotech (Garín) | Vaccine manufacturing, fermentation and cell culture, regulated | A and D |
| Gador, Laboratorios Bagó, Roemmers, Richmond, Biosidus, Amega/Gema Biotech | Pharma manufacturing and biotech in Buenos Aires province | D (data integrity) and A |
| Terragene (Rosario) | Biological indicators for sterilization, sells worldwide, lab-heavy QC | C and D |
| Porta Hnos, Bio4, ACA Bio, Ledesma | Industrial fermentation (ethanol) at scale, plants full of sensors and thin on analytics | A |
| Y-TEC (YPF) | Applied research, lithium and biotech, budgets | A and C |
| Mastellone, Quilmes/AB InBev, Peñaflor | Dairy, brewing, wine: fermentation with QC labs | A (yeast, fermentation curves) |
| Uruguay: ATGen, Genlives · Brazil: Biomm, Eurofarma, Aché, Cristália, Bio-Manguinhos, Butantan, Raízen · Chile: Andes Biotechnologies · Mexico: Landsteiner, Sophia, Probiomed | Regional expansion once the case studies exist (month 6+) | Same offers, in Spanish and Portuguese |

**Funds and programs as channels:** GridX (you already have a contact), SF500 (Rosario), CITES (Sunchales), Draper Cygnus, Kamay Ventures (food/agri corporate VC). Offer: "portfolio data-readiness package", 5 startups × 1-week mini audit, one consolidated report for the fund's partners.

**Scientific-data and bioprocess-AI companies (inspiration now, employer track later, verify each):** TetraScience, Uncountable, Benchling, Scispot, Sapio Sciences, LabKey, Synthace, DataHow (Switzerland, hybrid modeling for bioprocess), Novasign (Austria), Basetwo AI, Aizon (Barcelona, AI for pharma manufacturing, Spanish-speaking), Invert, Quartic.ai, Culture Biosciences, Ginkgo Bioworks, Cradle, LatchBio, Sartorius/Umetrics (SIMCA), Cytiva, Novonesis, DSM-Firmenich, Lonza, Novo Nordisk, NNE. These are the people who will read your preprints. Instrument 1 is, on purpose, a small open version of what DataHow and Aizon sell.

**Not on the list on purpose:** Big Four and large IT consultancies' life-sciences practices. Same culture you are leaving, different logo.

### 3.3 Culture filter (use it on every client and employer)

Ask, before signing anything:

1. Who decides priorities: scientists, or a delivery office?
2. What happened the last time a milestone slipped? (Listen for "we re-planned" versus "we worked weekends".)
3. How long is one experimental cycle here? (If the answer is in days, the company is paced by biology. Good.)
4. Is there on-call, and for what?
5. How many people left in the last 12 months, and why?
6. Can I see the last two retrospectives or lab meeting notes?

Two bad answers out of six: walk away. Your friend at Sumble is your calibration point: ask them to answer these six about their own workplace so you know what "good" sounds like.

---

## 4. Offers and pricing hypotheses

These prices are hypotheses to test with the first five conversations, not facts.

| Offer | What it is | Duration | Price hypothesis | Who buys |
|---|---|---|---|---|
| **Loop Audit** | Map the experimental loop, inventory data sources, measure cycle time and data readiness, run one instrument on their sample data, deliver 3 automation candidates with estimated days saved and one working prototype | 2 weeks, ~20 h of your time | USD 2,500 (seed startup, in exchange for a public anonymized case study) · USD 6,000-10,000 (Series A+, established, CDMO) | Founders, Head of R&D, QA managers |
| **Portfolio package** | 5 mini audits + one consolidated report for the fund | 5 weeks | USD 12,000-18,000 | Funds and programs |
| **Pilot** | One instrument adapted to their data and running in their lab, with a measured before/after | 6-8 weeks | USD 12,000-30,000 | Anyone who did an audit |
| **Fractional AI lead for R&D** | 1-2 days/week, owns the data-to-decision layer | Monthly retainer | USD 4,000-8,000/month | Companies with 2+ instruments running |

What the "burn to grow" comment implies for the pitch: never say "save money". Say "one fewer week per cycle × N cycles a year = M months of runway you did not have to raise". For pharma, say "batches not lost, findings not written".

Fixed price, half upfront, paid by invoice or card. No hourly billing in the first year: hourly reproduces EY's logic inside your own business.

---

## 5. The three instruments (plus the phase-2 backlog)

Each instrument follows the same seven steps, and the steps are the cure for the "instrument trap":

1. Build on **public data**.
2. Publish the **number** it produces (a deep dive, a DOI, a preprint).
3. Run it **for free on one founder's sample** (30 minutes of their time, NDA if needed).
4. Publish the **anonymized before/after**.
5. Put it inside a **priced Loop Audit**.
6. Ask the client for a **quote** you can publish.
7. Move on. Never add a feature that no client asked for.

### 5.1 Instrument 1 · `batchlens` — fermentation batch intelligence (weeks 1-8)

- **Gap it closes (archetype A):** paper or spreadsheet batch records, no golden batch, deviations detected at harvest, yield optimized by intuition.
- **Public data:** IndPenSim (industrial-scale penicillin fermentation simulator; the public release has 100 batches with online process variables and Raman spectra, including faulty batches). Verify the download link on industrialpenicillinsimulation.com. Plus your own Monod fed-batch simulator streamed over MQTT so the IoT gap is demonstrable without hardware.
- **What it does:** ingest batch CSVs or extracted paper records → batch unfolding and alignment → golden-batch envelope → multivariate monitoring (PCA, Hotelling T², SPE) → soft sensor for biomass and product titer from online variables → early deviation alarms → Bayesian optimization suggestion for the next run's feed profile → one HTML report per batch. `--fake` mode runs everything with no data.
- **Stack:** Python, pandas, scikit-learn, LightGBM, statsmodels, BoTorch (CPU is fine), Plotly, Mosquitto + Node-RED + InfluxDB + Grafana on a Debian box, optional ESP32 + temperature probe (USD 15) so one real sensor exists in the demo.
- **Hardware fit:** all models train on CPU or in seconds on the RTX 3050. The 6 GB card matters only in week 5 (vision model for batch-record extraction).
- **Headline numbers you publish:** hours-to-detect a deviation versus harvest time; titer prediction error at 50% of batch time; simulated titer gain after 12 BO-suggested runs versus a DoE grid.
- **Companies to show it to:** Michroma, Levya, Fungilife, NanoInGreen, POLYMERA, Hiamet, Food4You. Peers to ask for review: Mavios (they sell the enterprise version of this), Stämm.
- **Preprint 1 (week 8):** "Early deviation detection and soft sensing on an open industrial fermentation benchmark with consumer hardware: a reproducible baseline". arXiv cs.LG or q-bio.QM (needs an endorser), or bioRxiv (no endorser needed). Zenodo DOI for the code.

### 5.2 Instrument 2 · `trialkit` — from trials and assays to a defensible decision (weeks 9-13)

- **Gap it closes (archetypes B and C):** field trials and plate assays scattered across spreadsheets, subjective scoring, reports built by hand, no confidence intervals, so investors and growers do not trust the efficacy claim.
- **Public data:** field module: `agridat` (hundreds of published agricultural trials, CSVs available through the Rdatasets mirror) and Genomes to Fields; plate module: published growth-curve and dose-response datasets, PlantVillage for image scoring; Zindi/CGIAR agro challenges for extra practice.
- **What it does:** LLM-assisted mapping of any spreadsheet into one trial schema (plot, block, treatment, replicate, date, variable) → design detection (RCBD, split-plot) → mixed model with treatment effects and intervals → plate module: growth curves (logistic, Gompertz), 4PL dose-response, MIC, Michaelis-Menten, Z′-factor QC → image scoring (zero-shot CLIP, then a small fine-tuned classifier) → one report in Spanish or English that a grower, a fund or a regulator can read.
- **Stack:** pandas, statsmodels (MixedLM, ANOVA), scipy curve fitting, PyTorch small CNN or CLIP on the 3050, Claude or a local Qwen for schema mapping, the Vara harness to measure mapping accuracy and cost.
- **Headline numbers:** days from last measurement to report (manual versus trialkit); mapping accuracy of spreadsheets to schema; agreement between image scoring and human scoring.
- **Companies to show it to:** Puna Bio, Nunatak, Bsafe, Semion (field); Resistia, APEXzymes, Antarka, Syocin (plate). Peers: Beeflow, Zavia.
- **Output (week 13):** Medium deep dive 2, Zenodo DOI, 60-s video, **Loop Audit offer launched**.

### 5.3 Instrument 3 · `labdocs` — on-prem document intelligence with an audit trail (weeks 14-18)

- **Gap it closes (archetype D, and every archetype's "records on paper" line):** SOPs, protocols, batch records and notebook pages that are not searchable, not traceable and not auditable; agencies (ANMAT, FDA, EMA) demand ALCOA+ traceability the startup does not have.
- **Public data:** synthetic batch records and SOPs you generate from real templates (never client data), plus public regulatory documents. eLabFTW (open-source electronic lab notebook) running on your Debian box as the substrate, so the demo is "AI inside a real ELN", not a toy.
- **What it does:** extract structured experiments from PDFs and scanned pages with a local vision model (Qwen2.5-VL 3B or Gemma 3 4B on the 3050) → write to eLabFTW through its API with a hash-chained audit trail → local search and Q&A with citations (embeddings on the Debian box) → deviation report drafting from batchlens alarms plus the relevant SOP, human-in-the-loop → calibrated evaluation: 60 human-labeled cases, judge-human agreement published, local versus cloud accuracy and cost table.
- **Stack:** Ollama or llama.cpp (Qwen2.5 7B Q4, Qwen2.5-VL 3B, Gemma 3 4B, bge-m3 or nomic embeddings), eLabFTW in Docker, Postgres, the Vara judge, Claude and Gemini only as the cloud comparators.
- **Hardware fit:** this is where the small boxes become the argument. Publish the table: what runs on a USD 300 mini PC, what needs the 3050, what needs the cloud, with accuracy and cost per 1,000 pages.
- **Headline numbers:** extraction accuracy per field (local versus cloud); judge-human agreement; minutes from scanned page to auditable record.
- **Companies to show it to:** Eywa Biotech, AVaTAR MedTech, NanoproX, Synthetic Ocular Restoration, SphereBio, Neocell, plus any pharma QA manager.
- **Preprint 2 (week 18):** "On-premise document intelligence for GxP laboratories: a calibrated evaluation of small vision-language models on synthetic batch records and SOPs". arXiv cs.CL or cs.DL, or a JOSS submission for the software.

### 5.4 Phase-2 backlog (do not start before week 20)

- **Instrument 4 · clinical report automation** (archetype E: Enteria, Microgenesis, PreGen, CASPR, Ubique). Higher ticket, regulated (HIPAA/GDPR/Ley 25.326), needs a partner clinic. Start only if a diagnostic company asks after seeing instrument 3.
- **Scripts-to-product engineering** for archetype F companies with software gaps (Geoprot, ArgenTag, Monte Caldera, Zavia). Off-thesis but lucrative; take one only as a paid engagement, never as a portfolio project.

---

## 6. Theory catalog (what you study, why, and when)

All free. Roughly 40 hours over 20 weeks, two 1.5-h blocks per week. Each item names the practice it feeds.

| # | Piece | Source | Hours | Weeks | What it buys you |
|---|---|---|---|---|---|
| 1 | FDA guidance "PAT: A Framework for Innovative Pharmaceutical Development, Manufacturing and Quality Assurance" (2004) | fda.gov, PDF | 1.5 | 0 | The vocabulary of the people who sign pharma budgets: PAT, CPP, CQA, design space |
| 2 | Bioreactors / bioprocess engineering: growth kinetics (Monod), batch vs fed-batch mass balances, oxygen transfer (kLa), scale-up | NPTEL (IIT Madras "Bioreactors" lectures; verify current playlist) | 6 | 1-3 | Speak fermentation with a founder. Feeds the simulator and batchlens |
| 3 | NIST/SEMATECH e-Handbook of Statistical Methods, chapter 6 (process monitoring: Shewhart, EWMA, multivariate control charts) | itl.nist.gov/div898/handbook | 3 | 1, 4 | Golden batch and alarms with statistics, not thresholds |
| 4 | IndPenSim papers (Goldrick et al., 2015 J. Biotechnol.; 2019 Comput. Chem. Eng.) | open PDFs / publisher | 2 | 2 | Understand the dataset you build on |
| 5 | Multivariate statistical process control: PCA, PLS, Hotelling T², SPE for batch processes (any open tutorial; the SIMCA "MSPC" whitepapers are the industry reference) | open | 2 | 2, 4 | The method every bioprocess data scientist is expected to know |
| 6 | Hyndman & Athanasopoulos, Forecasting: Principles and Practice (3rd ed.), chapters 1-2, 5, 7 | otexts.com/fpp3 | 3 | 3 | Soft sensors are forecasting with leakage risk; learn the validation traps |
| 7 | CS336 lecture 12 (evaluation) + CS224N benchmarking lecture (carry-over from v3) | Stanford, YouTube | 2.5 | 5, 15 | The calibrated judge. Your signature move, applied to lab documents |
| 8 | Vision-language models for documents: Qwen2.5-VL and Gemma 3 model cards, Hugging Face "document AI" material | HF | 1.5 | 5 | Batch records on paper → JSON, locally |
| 9 | Frazier, "A Tutorial on Bayesian Optimization" (arXiv:1807.02811) + BoTorch tutorials | arXiv, botorch.org | 4 | 6 | "Which experiment next". The core of self-driving-lab thinking, runs on CPU |
| 10 | von Stosch et al., "Hybrid semi-parametric modeling in process systems engineering" (2014) | publisher / open preprint | 2 | 7 | Mechanistic + ML: the approach DataHow and Novasign sell |
| 11 | NIST e-Handbook chapter 5 (design of experiments: full/fractional factorial, RCBD, response surface) | itl.nist.gov | 3 | 9 | Read any trial design; design your own |
| 12 | Mixed models for designed experiments: statsmodels MixedLM docs + Piepho, Büchse & Emrich (2003) "A hitchhiker's guide to mixed models for randomized experiments" | docs, journal | 3 | 10 | The statistics a grower's agronomist or a regulator will accept |
| 13 | Assay statistics: 4PL dose-response, growth-curve models, Michaelis-Menten fitting, Z′-factor (Zhang et al., 1999) | open | 2 | 11 | Plate module of trialkit |
| 14 | Small image models: CLIP zero-shot, fine-tuning a ResNet/ViT on 6 GB (PyTorch tutorials) | pytorch.org | 2 | 12 | Replace subjective visual scoring |
| 15 | 21 CFR Part 11 + MHRA "GxP Data Integrity Guidance and Definitions" (2018, ALCOA+) + ICH Q10 (skim) | fda.gov, gov.uk, ich.org | 3 | 14 | What "audit trail" legally means. Nobody in the startups has read these |
| 16 | eLabFTW documentation and API | doc.elabftw.net | 1 | 14 | Real ELN substrate for labdocs |
| 17 | Retrieval evaluation and RAG failure modes (your own Vara notes + CS224N) | own | 1.5 | 15 | "Ask the lab" that can be trusted |
| 18 | Privacy and data governance basics: HIPAA, GDPR, Argentine Ley 25.326 (skim) | open | 1.5 | 17 | Needed for archetype E in phase 2 and for any clinical client |
| 19 | Science literacy thread (section 8): Rosalind, ESM-2, RDKit + TDC, scanpy, 16S basics | open | 8 | 1-20 | Talk to the AI-native companies as a peer |

Deferred, and why: Karpathy and CS336 internals (you learn them when an instrument needs them, not before), MIT 6.874 "Deep Learning in the Life Sciences" (excellent, 3 lectures allowed inside the literacy thread), full DoE or biostatistics courses (the NIST handbook covers what you need), protein design (Zavia and Nanogrow's territory; you are not competing there).

---

## 7. Week-by-week plan

Rhythm (same as v3, it fits your week): Tue and Thu theory (1.5 h each), Mon/Wed/Fri code (1.5 h each), Sat code (1.5 h, first to fall in a bad week), Sun publish (1 h). The Thursday block ends with 30 minutes of the science literacy thread or reading one founder's paper.

Legend: **G** = GitHub, **L-Sun** = LinkedIn Sunday build log, **L-Wed** = LinkedIn Wednesday gap note (archetype, never a named weakness), **M** = Medium, **X** = arXiv/bioRxiv, **Z** = Zenodo DOI, **V** = 60-s video.

### 7.1 Phase 0 · Positioning (week 0)

| Week | Lecture | Practice | Gap addressed | Public output |
|---|---|---|---|---|
| 0 | FDA PAT guidance (1.5 h). Read the founder pages of 10 archetype-A companies | Create the GitHub org/repo `labloop` with a 10-line README per instrument. Pull local models (Qwen2.5 7B Q4, Qwen2.5-VL 3B or Gemma 3 4B, bge-m3). On the Debian box: Docker with Mosquitto, Node-RED, InfluxDB, Grafana, eLabFTW. New LinkedIn headline and About (section 9.6) | Positioning | **L-Sun** post 0: "20 weeks building open instruments for wet labs, in public, at 1-2 h/day. First: fermentation." **G**: repo skeleton |

### 7.2 Phase 1 · Instrument 1, batchlens (weeks 1-8)

| Week | Lecture | Practice | Gap addressed | Public output |
|---|---|---|---|---|
| 1 | NPTEL: growth kinetics, batch and fed-batch balances. NIST ch. 6.1-6.3 (control charts) | Monod fed-batch simulator in ~80 lines (biomass, substrate, product, DO). Publish it over MQTT to the Debian box; Grafana dashboard with EWMA control limits. Optional: ESP32 + DS18B20 as one real signal | A: manual monitoring, no telemetry (Michroma, Levya, RADIAL, algaebio+) | **G** `batchlens/sim`. **L-Sun**: "A fermentation in 80 lines, streamed like a plant". **L-Wed**: gap note on the cost of a blind bioreactor |
| 2 | IndPenSim papers. MSPC tutorial (PCA on unfolded batches) | Load the 100 IndPenSim batches; batch unfolding and alignment; PCA on the golden batches; T² and SPE limits; plot the faulty batches against the envelope | A: no golden batch | **G** `batchlens/mspc`. **M1** thesis piece: "Biotech startups do not have an AI problem" (1,500 words, the archetypes with counts, no names). **L-Sun** + **L-Wed** |
| 3 | NPTEL: oxygen transfer, scale-up. FPP3 ch. 5 and 7 (regression, validation) | Soft sensor: predict biomass and penicillin concentration from online variables; batch-wise cross-validation (no leakage across batches); LightGBM versus PLS baseline; error by batch time | A: no prediction of yield mid-batch (Fungilife, NanoInGreen, Bioplastix) | **G** `batchlens/softsensor`. **L-Sun**: "How early can you know the titer? Error at 25/50/75% of batch time". **L-Wed** |
| 4 | NIST ch. 6.5 (multivariate charts). MSPC for fault detection | Early deviation detection on the IndPenSim faulty batches; metric = hours-to-detection versus harvest; false-alarm rate; a `batchlens monitor` command that replays a batch and raises alarms | A: deviations found at harvest (Neocell, Levya, Protiva) | **G**. **L-Sun** with the detection chart. **L-Wed** |
| 5 | CS336 L12 (evaluation). Qwen2.5-VL / Gemma 3 model cards | Generate 200 synthetic paper batch-record pages (templates with handwriting fonts, stains, rotations). Extract to JSON with a local 3B VLM on the 3050 versus Claude and Gemini. Reuse the Vara harness: accuracy per field, cost, latency. 60 pages hand-checked | A and D: batch records on paper (Levya, APEXzymes, Eywa) | **G** `batchlens/records`. **L-Sun**: "Paper batch records to JSON: a 3B local model versus the cloud, with the table". **L-Wed** |
| 6 | Frazier BO tutorial. BoTorch "closed loop" tutorial | Bayesian optimization of the feed-rate profile in the simulator (4-6 parameters, titer objective, contamination-risk constraint). Compare 12 BO runs against a 12-point DoE grid. Then run it on IndPenSim-fitted surrogate | A: yield optimized by intuition | **G** `batchlens/nextrun`. **L-Sun**: "12 simulated runs, +X% titer: what Bayesian optimization does for a fermentation". **L-Wed** |
| 7 | von Stosch hybrid modeling review | Package: `batchlens report batch.csv` → HTML (golden envelope, alarms, soft-sensor curve, next-run suggestion). `--fake` mode. README with the numbers. Ask your Sumble friend for a 30-min code review (a senior architect's review is free and rare) | A, whole loop | **G** release v0.1 + **Z** DOI. **L-Sun**: the report. **L-Wed** |
| 8 | None (writing week) | Preprint 1 draft using the NotebookLM prompt 6 skeleton (carry-over). Ask one founder-scientist for endorsement or co-authorship. Send the report to 5 archetype-A founders asking for 30 minutes of feedback (not selling) | Founder feedback loop | **X1** submitted. **M2** deep dive: "Early deviation detection on 100 open fermentation batches with a laptop GPU". **V1**. **L-Sun** |

### 7.3 Phase 2 · Instrument 2, trialkit, and the first invoice (weeks 9-13)

| Week | Lecture | Practice | Gap addressed | Public output |
|---|---|---|---|---|
| 9 | NIST ch. 5.1-5.3 (DoE). agridat overview | Trial schema. LLM-assisted mapping of 10 heterogeneous agridat spreadsheets to the schema (Claude versus local Qwen); mapping accuracy measured with the harness | B: trials scattered in spreadsheets (Puna Bio, Nunatak, Bsafe) | **G** `trialkit/ingest`. **L-Sun**: "10 field trials, 10 spreadsheet formats, one schema". **L-Wed** |
| 10 | Mixed models (statsmodels MixedLM; Piepho guide) | Automated analysis: design detection, ANOVA and mixed model, treatment effects with intervals, bootstrap; `trialkit report` in Spanish and English. Incorporate feedback from week 8 conversations into batchlens (one afternoon, no more) | B: no defensible efficacy report for growers and investors (Semion, Nunatak) | **G**. **L-Sun**: "The report a grower would believe". **L-Wed** |
| 11 | Assay statistics: 4PL, growth curves, Michaelis-Menten, Z′-factor | Plate module: parse plate-reader exports (96/384 layouts), growth curves, MIC, enzyme kinetics, dose-response with intervals, plate QC | C: assays analyzed by hand in Excel (Resistia, APEXzymes, Antarka, Syocin) | **G** `trialkit/plate`. **L-Sun** + **L-Wed** |
| 12 | Small image models (CLIP zero-shot, fine-tune on 6 GB) | Image scoring on PlantVillage: zero-shot versus a fine-tuned small model on the 3050; agreement with labels; a `trialkit score-images` command. Write the **Loop Audit one-pager** (section 9.7) | B: subjective visual scoring (Semion, Bsafe) | **G**. **L-Sun**: "Replacing the eye: image scoring on a 6 GB GPU". **L-Wed** |
| 13 | None (packaging and selling week) | trialkit v0.1 release. Offer sent to 10 people: 5 founders from archetypes A-C who replied in week 8, 3 established companies, 2 funds. Goal: 3 calls booked | The charging muscle | **G** + **Z** DOI. **M3** deep dive: "From spreadsheet to defensible efficacy claim: trialkit on N published trials". **V2**. **L-Sun**: the offer, stated plainly. **L-Wed** |

**Gate at week 13:** if fewer than 3 calls are booked, week 14 becomes a second selling week and instrument 3 shifts by one week. The calendar bends; the gate does not.

### 7.4 Phase 3 · Instrument 3, labdocs (weeks 14-18)

| Week | Lecture | Practice | Gap addressed | Public output |
|---|---|---|---|---|
| 14 | 21 CFR Part 11, ALCOA+, ICH Q10. eLabFTW API | eLabFTW on the Debian box. Pipeline: PDF/scan → local VLM → structured experiment → eLabFTW entry with a hash-chained audit log. 200 synthetic SOPs and protocols. **First Loop Audit delivered or in progress** | D: no digital audit trail (Eywa, AVaTAR, NanoproX) | **G** `labdocs/ingest`. **L-Sun**: "What ALCOA+ means in code". **L-Wed** |
| 15 | Retrieval evaluation; CS224N benchmarking (judge bias) | "Ask the lab": local embeddings on the Debian box, citations to the exact protocol paragraph; 60 human-labeled Q&A; judge calibration; agreement number published | D and every archetype's "not searchable" line | **G** `labdocs/ask`. **L-Sun**: "Judge-human agreement: X%. Everything else depends on that". **L-Wed** |
| 16 | Light: agent patterns for human-in-the-loop drafting | Deviation report drafting: batchlens alarm + relevant SOP → CAPA/deviation draft → human approval → eLabFTW. **First invoice sent by this week** (gate) | A + D: deviations documented late (Eywa, Neocell, any pharma QA) | **G** `labdocs/deviation`. **L-Sun**: the anonymized audit case study, with the client's permission. **L-Wed** |
| 17 | Privacy and data governance basics | Hardening: Docker compose of the whole stack on a mini PC; benchmark table: USD 300 box versus 3050 versus cloud (accuracy, cost per 1,000 pages, minutes per page) | The on-prem argument for IP-sensitive labs | **G** + **Z** DOI. **L-Sun**: "A lab's AI on a USD 300 computer". **L-Wed** |
| 18 | None (writing week) | Preprint 2 draft. JOSS submission optional. Second audit in progress if sold | | **X2** submitted. **M4** deep dive. **V3**. **L-Sun** |

### 7.5 Phase 4 · Consolidate and go outbound (weeks 19-20)

| Week | Lecture | Practice | Public output |
|---|---|---|---|
| 19 | None | One-screen personal site: three instruments with their numbers, two preprints, the case study, the offer. 20 targeted messages (10 startups, 5 established/CDMO, 3 funds, 2 scientific-data companies). Employee track in parallel: 5 applications to scientific-data or bioprocess-AI companies with the preprints as the cover letter | **L-Sun** + **L-Wed** |
| 20 | NotebookLM mock interview (carry-over prompt 3) | Talk proposal: GridX community, a biotech meetup in Buenos Aires or Rosario, or an online LATAM one: "What 100 open fermentation batches and 30 published field trials taught me about the biotech data loop". Public retro. Plan months 6-12 (phase-2 backlog, retainers) | **L-Sun** retro. Talk submitted |

---

## 8. Science literacy thread (30-45 min on Thursdays, inside the theory hours)

You cannot sell to scientists without reading their language. This thread is small, weekly, and produces one Wednesday gap note per month.

| Weeks | Topic | Practice | Why (who it lets you talk to) |
|---|---|---|---|
| 1-4 | Molecular biology basics through Rosalind (first 12 problems) + one founder paper per week | Solve in Python, commit | Any founder. Most are CONICET scientists; reading their paper is the best outreach line you have |
| 5-8 | Protein language models: ESM-2 (650M fits the 3050 in fp16) embeddings, enzyme family classification on a small public set; MIT 6.874 protein lecture | One notebook | APEXzymes, Antarka, Resistia, Zavia, Nanogrow |
| 9-12 | Molecules: RDKit fingerprints, one ADMET task from Therapeutics Data Commons with a gradient-boosting baseline; dmol.pub GNN chapter | One notebook | Scitherm, Dogma, CellCo, Arakion |
| 13-16 | Single-cell basics with scanpy on a small PBMC dataset; MIT 6.874 single-cell lecture | One notebook | ArgenTag, OncoPrecision, CellRep |
| 17-20 | Microbiome basics: 16S concepts, a public amplicon dataset, diversity metrics | One notebook | Enteria, Microgenesis, Puna Bio, Hiamet |

Four notebooks in 20 weeks. They are literacy, not instruments: they never get a release, a DOI or a feature.

---

## 9. Publishing engine

### 9.1 Channels and cadence

| Channel | Format | Cadence | Real objective |
|---|---|---|---|
| GitHub (`labloop` org) | Three instrument repos, each with README numbers, `--fake` mode, `make report`, Zenodo DOI | Commits Mon/Wed/Fri/Sat | The proof. The only thing a senior engineer or a CTO actually checks |
| LinkedIn | Sunday build log (template A). Wednesday gap note (template E) | 2 per week | Founders, R&D heads, QA managers, fund partners, recruiters. Your management past is an asset here |
| Medium | Deep dive, 1,500-2,500 words, one chart that carries the argument | Weeks 2, 8, 13, 18 | What people read before they write to you. Cross-post to LinkedIn articles |
| arXiv / bioRxiv | Preprint with reproducible code and DOI | Weeks 8 and 18 | Credibility with scientists and with the scientific-data companies. Note: arXiv needs an endorser for a first submission in a category; bioRxiv does not. Ask a founder-scientist in week 6 |
| Zenodo | DOI per release | Weeks 7, 13, 17 | Citable software. Free, instant |
| Video | 60 seconds, terminal only (template C) | Weeks 8, 13, 18 | Founders watch, they do not read READMEs |
| X | Copy of the Sunday build log, more technical | Optional, 5 min | The bioprocess-AI and scientific-data people live there |
| Talk | 20 minutes | Proposed week 20 | Converts the whole program into one conversation with a room |

Rules:

- The Wednesday note is derived from the week's work, 10 minutes, two paragraphs. If it takes 30, skip it.
- Every post has one number or one admitted failure. Never "the ultimate guide".
- Archetypes, never named weaknesses. Name a company only to praise or after they engaged.
- Zero EY data, names, clients or comparisons. All code on your time and hardware, commits dated.

### 9.2 Template A · Sunday build log (20 minutes)

```
Week [N] of 20, building open instruments for wet labs at 1-2 h/day.

This week: [ONE thing built or measured]
The number: [the measurement, with N and an interval when you have one]
Where it fails: [one honest line]
Next: [one thing]

[image: chart, table or terminal capture]
Repo: [link]
```

### 9.3 Template E · Wednesday gap note (10 minutes)

```
A [archetype description: "fungal fermentation startup", "bioinput company running field trials"] typically [the gap, stated as a process fact, not a fault].

The cost is not the spreadsheet. It is [the days lost, the batch lost, the claim an investor does not believe].

An open way to close it: [what the instrument does, one sentence] → [link to the week's commit or chart].

If this is your loop, I would like to hear how you do it today. No pitch.
```

### 9.4 Template B · Deep-dive and preprint titles that work

```
1. "Early deviation detection on 100 open fermentation batches: [X] hours before harvest, [Y]% false alarms"
2. "How early can you predict titer? Soft-sensor error at 25/50/75% of batch time on IndPenSim"
3. "Paper batch records to JSON: a 3B local model versus the cloud, per-field accuracy and cost per 1,000 pages"
4. "From spreadsheet to defensible efficacy claim: [N] published field trials, one schema, one report"
5. "On-premise document intelligence for GxP labs: judge-human agreement [X]% (n=60)"

Rule: the title carries a measured number or an admitted failure.
```

Preprint skeleton (both preprints): problem in days and money → data and why it is public → method, reproducible in one command → results with intervals → where it fails → limitations of the judge or the metric → what you did not test. Mark every unsupported claim [VERIFY] before submission.

### 9.5 Template C · 60-second video

```
0-5 s: the problem, in days or in a lost batch.
5-15 s: one command running. Camera on the terminal.
15-45 s: the chart appearing. No cuts.
45-55 s: where it fails. One sentence.
55-60 s: the link.
No intro, no music, large terminal font.
```

### 9.6 Profiles (week 0)

LinkedIn headline:
`Applied AI for biotech R&D and biomanufacturing | I turn lab and plant data into weekly decisions | Open instruments: batchlens · trialkit · labdocs | 25 yrs leading delivery, ex-Big Four AI transformation lead`

About (5 lines): the thesis in one sentence; the three instruments with one number each as they appear; the offer in one line; "everything is open, measured, and reproducible on a USD 300 computer"; link to the repo and the calendar.

GitHub: real photo, one-line bio with the thesis, `labloop` pinned, profile README with the three numbers.

### 9.7 Template F · Loop Audit one-pager (week 12)

```
LOOP AUDIT — 2 weeks, fixed price

What you get
- A map of your experimental loop: every step from hypothesis to decision, with days per step.
- Your two numbers: loop cycle time and data readiness (% machine-readable within 24 h).
- Inventory of every data source (instruments, spreadsheets, paper, ELN) and where the days go.
- Three automation candidates ranked by days saved per cycle, with effort estimates.
- One working prototype on a sample of your data, running on your machine or mine.
- A 45-minute readout with your team.

What I need from you
- 3 hours of interviews (founder, lab lead, whoever keeps the spreadsheet).
- A sample of real data under NDA, or synthetic data if you prefer.

Price: USD [X], half upfront. Case-study price for seed-stage companies in exchange for an anonymized write-up.
```

### 9.8 Template D · Targeted message (weeks 8, 13, 19)

```
Hi [NAME],

I read [their paper / a specific thing about their process]. [One sentence showing you understood it.]

I build open instruments for wet-lab data loops. The last one, [instrument], [one number: "detects fermentation deviations X hours before harvest on 100 open batches" / "turns 10 spreadsheet formats into one defensible efficacy report"]. Code, data and a preprint: [link].

Before this I led AI transformation at a Big Four firm, so I care about the invoice as much as the model.

If [archetype gap] is part of your week, I would like 30 minutes to hear how you do it today. If it is useful, I can run it on a sample of your data, no cost, no pitch.

[NAME]
```

---

## 10. Hardware, software and budget

| Resource | Role in the plan |
|---|---|
| RTX 3050 6 GB + 32 GB RAM + Ryzen | All training (GBM, small CNNs, PLS, BO) and local 3-7B models (Q4). ESM-2 650M in fp16. Enough for everything in weeks 1-20 |
| Debian mini boxes | The on-prem story: Mosquitto, Node-RED, InfluxDB, Grafana, eLabFTW, Postgres, embeddings and a 1-2B model. Week 17 benchmarks them explicitly |
| Ollama / llama.cpp / LM Studio | Qwen2.5 7B Q4, Qwen2.5-VL 3B or Gemma 3 4B (vision), bge-m3 or nomic-embed |
| Claude, OpenAI, Gemini, OpenRouter | Cloud comparators in the accuracy/cost tables; Claude for schema mapping and judge; Gemini for long documents |
| NVIDIA free API | Large open models (Nemotron, Llama 70B) as a third comparator at zero cost |
| Google Colab free | Only if ESMFold or a bigger model is needed in the literacy thread |
| Hugging Face | Models, and a dataset card for the synthetic batch records and SOPs (week 17) |
| Optional ESP32 + DS18B20 | USD 15. One real sensor in the demo makes the IoT conversation concrete |

Budget: APIs USD 150-200 over 20 weeks (extraction comparisons in weeks 5 and 15 are the only expensive runs; everything else is `--fake` or local). Hardware USD 0-15. Everything else is free.

---

## 11. Metrics and gates

Weekly board (30 seconds on Sunday):

| Metric | Target | Why |
|---|---|---|
| Blocks completed | 5 of 7 | The only effort metric |
| Days with a commit | 3 of 7 | Progress, not video hours |
| Sunday build log | 1 | The forcing function |
| Wednesday gap note | 1 (skippable) | Keeps the audience oriented to the gaps |
| Founder conversations (cumulative) | 5 by week 9, 15 by week 16, 30 by week 20 | The strategy's leading indicator |
| Lost weeks in a row | Never 2 | The only hard limit |

Gates:

| Week | Gate | If missed |
|---|---|---|
| 8 | Preprint 1 submitted; 5 founders contacted | Week 9 becomes the writing week; trialkit shifts by one |
| 13 | Offer sent to 10; 3 calls booked | Week 14 is a second selling week |
| 16 | **First invoice sent** (any amount) | Stop building. Weeks 17-18 become selling weeks. This is the gate that matters |
| 20 | 2 preprints, 3 DOIs, 4 deep dives, 1 case study, 30 conversations, 1 talk proposal, 5 applications | Honest retro, then phase 2 |

Vanity metrics, banned: stars, likes, followers, hours of video, courses completed.

Honest expectation for week 20: evidence, one or two paid audits, a pipeline of conversations, and a clear view of which of the four archetypes pays fastest. The retainer or the job offer is the month 6-9 target.

---

## 12. Ignore for now

- **Vara as a flagship.** Do not start it. Its harness and judge live on inside batchlens (week 5) and labdocs (week 15).
- **Instrument 4 (clinical reporting)** before week 20, and any regulated clinical data at all in this phase.
- **Protein or molecule design.** Zavia, Nanogrow, CellCo own that. You do literacy, not competition.
- **Building a LIMS/ELN product.** eLabFTW exists. You add AI to it; you do not replace it.
- **A fourth instrument, a SaaS, a dashboard product, a marketplace.** Module 23 in a lab coat.
- **Hourly consulting.** Fixed price only in year one.
- **Any wet-lab work, any hardware beyond one sensor.** You are the data-to-decision layer, not the lab.
- **Full courses**: MIT 6.874 complete, CS336 complete, biostatistics MOOCs, cloud certifications. The catalog in section 6 is all that fits.
- **Naming a company's weakness in public.** Ever.
- **A second project.** El Espejo stays at its current low-effort pace and does not count.

---

## 13. Execution protocol

- **One block, one thing.** Theory or code, never both in the same block. Tue/Thu theory, Mon/Wed/Fri/Sat code, Sun publish.
- **Code first, theory when the code asks.** Try the practice before the lecture; go to the lecture with a concrete question.
- **Tomorrow's TODO** is the last line of every block, written in the repo.
- **Sunday hour is sacred:** build log (20 min), board (5 min), NotebookLM gap detector (20 min, carry-over prompt 5), TODO for the week (15 min).
- **Publish even when the week was bad.** "This week I did not advance; here is what blocked me" is a valid and well-received post.
- **Thursday's last 30 minutes** go to the literacy thread or to one founder's paper. Reading their paper is the outreach.
- **Boredom is a signal.** Three bored blocks on a topic means it is not connected to an instrument. Skip it.
- **Hygiene while employed:** zero EY data or names, own hardware, dated commits, no comparisons with your employer. Your narrative is what you build, never what your firm does wrong.

### Tomorrow's first block (60-90 minutes)

1. Create the `labloop` GitHub org and the `batchlens` repo with a 10-line README: what it measures, on which public dataset, when the first number ships (week 4).
2. Pull Qwen2.5 7B Q4 and one vision model with Ollama; confirm they load on the 3050.
3. Change the LinkedIn headline (section 9.6).
4. Post 0, five lines: "20 weeks building open instruments for wet-lab data loops, in public, at 1-2 h/day. First instrument: fermentation batch intelligence on 100 open batches. Week 0."
5. Write tomorrow's TODO: "Monod fed-batch simulator, 80 lines, publishing to MQTT."

---

## 14. The one thing to remember

In 20 weeks nobody in a lab will ask how many lectures you watched. A founder will ask: **how many days does your instrument take out of my loop, how do you know, and where does it fail?** This document exists so that you can answer with a link, a number and an invoice.

---

## Appendix A · Full company map (86 companies from the GridX list)

Archetype counts (primary fit): A · Process & telemetry: 25 · B · Field trials: 8 · C · Lab assays & preclinical: 14 · D · Traceability & QMS: 7 · E · Diagnostics & clinical reporting: 11 · F · AI-native / computational: 20 · Off-thesis: 1. AI/IoT base declared: 22 of 86.

The gap column is the private hypothesis from the CSV, shortened. It is never to be quoted in public.

### A · Process & telemetry — instrument: 1 · batchlens — role: Proof partner → paying candidate

| Company | AI/IoT base | What they do | Hypothesized gap (private) | Secondary fit |
|---|---|---|---|---|
| algaebio+ | no | Desarrollo y producción de compuestos beneficiosos a partir de microalgas | Monitoreo manual de estanques o biorreactores y falta de control predictivo sobre el crecimiento de biomasa |  |
| Algalife | no | Plataforma para la optimización de procesos productivos basados en microalgas | Desconexión entre el protocolo biológico y el control digital continuo de parámetros físico-químicos |  |
| Alkemio | no | Tecnología de reconocimiento molecular para separación de tierras raras de alta eficiencia | Falta de software de control de proceso industrial para la reproducibilidad de lotes a escala |  |
| BioBlends | no | Bioconservantes a base de gas para extender la vida útil de alimentos frescos | Pérdida de datos en cadena de frío por falta de telemetría y sensores conectados |  |
| BioPlaster Research Inc | no | Packaging biodegradable producido a partir de biomasa de algas invasoras (sargazo) | Procesos de manufactura semi-artesanales y baja trazabilidad de parámetros de extrusión y resistencia |  |
| Bioplastix | no | Uso de bacterias seleccionadas para sintetizar plásticos biodegradables de alto rendimiento | Desconocimiento de herramientas de automatización industrial para optimizar el rendimiento por tanda |  |
| BreakPET | no | Reciclaje y degradación de plástico PET mediante enzimas modificadas genéticamente | Carencia de software para modelar el balance de masa/energía y validar la viabilidad técnico-económica cont… |  |
| ByBug | no | Biofabricación terapéutica mediante insectos modificados criados como biorreactores vivos | Operaciones de cría intensivas en mano de obra con alta variabilidad biológica no monitorizada |  |
| Calfix | no | Biotecnología bacteriana para autorreparar y sellar fisuras en estructuras de hormigón | Inexistencia de un componente tecnológico digital que complemente la solución biológica en la obra civil |  |
| Caligenia | no | Biofertilizantes con huella de carbono negativa a partir del tratamiento de estiércol avícola | Falta de sistemas de automatización en la planta de tratamiento para garantizar uniformidad de producto |  |
| CyanoMin | no | Biorremediación de salmueras hipersalinas con recuperación de minerales estratégicos como litio | Falta de control automático de caudales y monitoreo continuo en ambientes industriales agresivos |  |
| Einsted | no | Transformación catalítica de metano y dióxido de carbono en energía limpia y nanoestructuras de carbono | Control manual de parámetros de proceso crítico en reactores químicos de alta temperatura |  |
| Enzyva | no | Biofloculante multi-enzimático para el tratamiento de efluentes de la industria textil | Falta de demostración empírica automatizada de reducción de contaminantes para clientes industriales |  |
| Food4You | no | Bacterias y consorcios microbianos para fermentar alimentos de origen vegetal | Variabilidad entre lotes por falta de control sensorizado y analítica de datos en producción |  |
| Fungilife | no | Sustitución de tensioactivos sintéticos por biosurfactantes derivados de hongos | Ausencia de una plataforma unificada para optimizar parámetros cinéticos del cultivo fúngico |  |
| Hiamet S.A | no | Biotecnología y consorcios microbianos para optimizar la producción de biogás | Operación ciega de digestores industriales sin datos predictivos para evitar la acidificación |  |
| Levya | no | Lípidos y aceites sustentables producidos mediante fermentación de levaduras modificadas | Escalado dependiente de intervención manual en biorreactores y registros dispersos en papel/hojas de cálculo |  |
| Michroma | no | Biofábricas fúngicas para producir colorantes e ingredientes naturales sostenibles | Procesos de laboratorio manuales y falta de infraestructura digital para optimizar el rendimiento del escal… |  |
| NanoInGreen | no | Ingredientes bioactivos naturales utilizando microorganismos como nanofábricas celulares | Carencia de herramientas analíticas para correlacionar parámetros de proceso con rendimiento de biomasa |  |
| Nat4Bio | no | Recubrimientos biológicos comestibles para protección postcosecha de frutas y verduras | Falta de validación continua en tiempo real de la tasa de degradación de fruta durante el transporte logístico | B |
| Neocell | no | Plataforma para expansión y producción a gran escala de células madre hematopoyéticas | Procesos de cultivo manuales propensos a contaminación y con escasa monitorización digital continua | D |
| POLYMERA | no | Superabsorbentes biodegradables a base de polímeros bacterianos antárticos | Producción en escala piloto sin control digital de variables operativas críticas |  |
| Protiva | no | Colorantes textiles biológicos generados mediante fermentación bacteriana | Procesos de tinción empíricos sin registro sensorizado de parámetros como pH, temperatura y fijación |  |
| RADIAL | no | Biomateriales y aislantes producidos a partir de micelio fúngico y rastrojos agrícolas | Pérdidas de producción por contaminación fúngica al carecer de control ambiental automatizado |  |
| Tomorrow | no | Ingredientes y sistemas de proteínas vegetales aisladas para la industria de alimentos | Falta de trazabilidad digital del abastecimiento de granos e inconsistencias de textura no sensorizadas |  |

### B · Field trials — instrument: 2 · trialkit (field) — role: Proof partner → paying candidate

| Company | AI/IoT base | What they do | Hypothesized gap (private) | Secondary fit |
|---|---|---|---|---|
| AQUIT | no | Reemplazo de antibióticos para la acuicultura y piscicultura | Dependencia de muestreo manual en campo y carencia de infraestructura para análisis de salud animal continuo |  |
| Bsafe Biotech | no | Productos basados en ARN de interferencia (ARNi) para proteger cultivos y animales | Ensayos agronómicos descentralizados sin plataforma común para consolidar indicadores de desempeño |  |
| FeedVax | no | Vacunas orales para peces que reemplazan la vacunación inyectable en acuicultura | Dificultad para registrar y medir la tasa de ingestión y protección efectiva en campo sin telemetría |  |
| MicroIN | no | Tecnología de microencapsulación de microorganismos para bioinsumos agrícolas | Falta de trazabilidad digital del producto durante el canal de distribución y logística en el agro |  |
| Nunatak Biotech | no | Bioinsumos para agricultura resistente a estrés biótico y abiótico a partir de flora andina | Registros en campo no estructurados que retrasan la validación comercial frente a productores y fondos |  |
| Puna Bio | no | Bioinoculantes agrícolas a partir de bacterias extremófilas para cultivos en condiciones adversas | Gestión experimental dispersa en hojas de cálculo y falta de digitalización sistemática de pruebas agronómicas |  |
| Qumir Nano | no | Nanotecnología aplicada al desarrollo de bioinsumos y fitosanitarios para el campo | Dificultades para demostrar consistencia lote a lote por falta de registros de producción digitalizados | D |
| Semion | no | Estimulación molecular para activar los mecanismos de defensa inmune en plantas | Validación de campo basada en evaluaciones visuales subjetivas sin analítica de datos automatizada |  |

### C · Lab assays & preclinical — instrument: 2 · trialkit (plate) — role: Proof partner → paying candidate

| Company | AI/IoT base | What they do | Hypothesized gap (private) | Secondary fit |
|---|---|---|---|---|
| Antarka | no | Enzimas antárticas para formulaciones de longevidad cutánea | Procesos de laboratorio no integrados a plataformas de datos y pruebas clínicas sin trazabilidad digital |  |
| APEXzymes | no | Enzimas de bajo costo para procesos industriales verdes y rentables | Ausencia de monitoreo en tiempo real de cinéticas y registro manual de experimentos de laboratorio | A |
| Arakion | no | Diseño racional de insecticidas a partir de toxinas del veneno de arácnidos | Flujo de trabajo netamente químico/biológico sin herramientas modernas de análisis de datos colaborativos |  |
| BioClé | no | Vesículas bacterianas para transformar la salud digestiva mediante entrega dirigida de activos | Registro manual de ensayos de formulación y falta de sistemas integrados de control de lotes |  |
| Bioeutectics | no | Solventes eutécticos naturales, verdes y biodegradables a partir de compuestos biológicos | Gestión manual de recetas químicas y ausencia de digitalización en el control de calidad de lotes | D |
| Invitrall | no | Modelos de órganos humanos in vitro para evaluación preclínica de eficacia y toxicidad | Reportes de toxicidad generados manualmente con riesgo de errores de transcripción y lentitud de entrega |  |
| Kresko RNAtech | no | Ingredientes de bienestar basados en microARN bioactivos de origen alimentario | Falta de herramientas de software para gestionar la complejidad de las dianas moleculares identificadas |  |
| Lipock | no | Nanocápsulas lipídicas para mejorar la estabilidad de activos volátiles e hidrofóbicos | Dificultad para correlacionar variaciones operativas del equipo de dispersión con la estabilidad final |  |
| Nanotransfer | no | Vectores no virales para administración y entrega eficiente en terapias génicas | Sistemas de laboratorio tradicionales con baja integración de datos analíticos para optimizar vectores |  |
| Resistia | no | Enzimas y soluciones biológicas para disgregar biofilms e infecciones multirresistentes | Ensayos de eficacia en microplacas no digitalizados, generando reportes lentos y dependientes de Excel |  |
| Scitherm Therapeutics | no | Moléculas pequeñas para activación metabólica del tejido adiposo pardo contra la obesidad | Fragmentación de datos de química médica que ralentiza la selección del compuesto líder |  |
| Syocin Bio | no | Bacteriocinas sintéticas de alta precisión para control selectivo de bacterias patógenas | Lenta iteración entre el diseño biológico y los ensayos de laboratorio por falta de integración de datos |  |
| Vexxel Biotech | no | Nanocápsulas de proteínas inteligentes para liberación controlada en agricultura y cosmética | Ensayos de liberación manuales sin correlación digital de variables ambientales como radiación o pH |  |
| Werken Vac | no | Vacunas de ARN autoamplificable para prevención de enfermedades virales en peces | Gestión de ensayos en jaulas o tanques sin sistemas de monitoreo continuo de inmunización |  |

### D · Traceability & QMS — instrument: 3 · labdocs — role: Paying candidate (regulatory pressure)

| Company | AI/IoT base | What they do | Hypothesized gap (private) | Secondary fit |
|---|---|---|---|---|
| Apasomics | no | Nanopartículas de hidroxiapatita que liberan bisfosfonatos dirigidos al tejido óseo | Falta de digitalización en el seguimiento de ensayos in vivo/in vitro y registros documentales descentraliz… |  |
| AVaTAR MedTech | no | Dispositivos médicos para la reconstrucción mínimamente invasiva de válvulas cardíacas | Carencia de plataformas de software para seguimiento post-comercialización y gestión de datos de ensayos cl… |  |
| Eywa Biotech | no | Plataforma biotecnológica para la biosíntesis precisa de principios activos psicodélicos | Cumplimiento regulatorio vulnerable por falta de pistas de auditoría digitalizadas en tiempo real | A |
| NanoproX | no | Nanopartículas de sulfuro de bismuto como radiosensibilizadores para potenciar radioterapia | Datos preclínicos fragmentados y falta de trazabilidad digital exigida por agencias regulatorias |  |
| New Organs Biotech | no | Edición genética y medicina regenerativa para resolver la escasez de órganos para trasplante | Manejo descentralizado de registros genéticos complejos sin infraestructura de auditoría robusta |  |
| SphereBio | no | Nanoplataformas derivadas de tumores para vacunas terapéuticas personalizadas contra el cáncer | Riesgo de errores en la personalización de vacunas por falta de un sistema de información centralizado |  |
| Synthetic Ocular Restoration | no | Biomateriales e ingeniería de tejidos para regeneración de córneas y patologías visuales | Protocolos de manufactura manuales sin integración de software para el cumplimiento regulatorio |  |

### E · Diagnostics & clinical reporting — instrument: 4 · phase 2 — role: Phase 2

| Company | AI/IoT base | What they do | Hypothesized gap (private) | Secondary fit |
|---|---|---|---|---|
| CASPR Biotech | yes | Diagnóstico molecular descentralizado y portátil basado en CRISPR para detección rápida de patógenos | Brecha entre el biosensor/química CRISPR y el software clínico comercial integrable con sistemas hospitalar… |  |
| Embryoxite | yes | Diagnóstico no invasivo para evaluar embriones y optimizar tratamientos de fertilización in vitro | Dependencia de procesos manuales de laboratorio para cargar datos en lugar de integración directa con equipos |  |
| Enteria | no | Desarrollo de herramientas de diagnóstico de salud de precisión basadas en la microbiota | Generación manual de reportes y carencia de infraestructura para escalar el volumen de análisis |  |
| Fecundis | no | Tecnologías para mejorar la funcionalidad del esperma y la eficiencia en reproducción asistida | Falta de integración tecnológica entre los insumos biológicos y los sistemas de información de las clínicas |  |
| Gisens Biotech | yes | Diagnósticos descentralizados ultraprecisos con biosensores basados en nanotecnología | Riesgo de diseñar un gran biosensor físico pero carecer de un ecosistema de software conectado y robusto |  |
| GlycoX | no | Píldora oral para estabilización y control de niveles óptimos de glucosa | Falta de una solución digital que demuestre la eficacia en el mundo real mediante métricas continuas |  |
| Microgenesis | no | Test diagnóstico del microbioma para identificar y tratar causas de infertilidad femenina | Flujo operativo manual desde la toma de muestra hasta el informe final enviado a los especialistas |  |
| NanoJump Bio | no | Sistemas portátiles de extracción de material genético mediante nanotecnología | Procesos de campo dependientes de registro manual del operador sin geolocalización ni sincronización |  |
| PreGen | no | Tests de diagnóstico y pronóstico genético de cáncer accesibles y de bajo costo | Inexistencia de un portal digital seguro para médicos y laboratorios que automatice la entrega diagnóstica |  |
| Ubique Bio | no | Plataforma descentralizada de pruebas diagnósticas rápidas utilizando péptidos sintéticos | Test rápido desarrollado solo a nivel reactivo químico sin plataforma digital para centralizar resultados |  |
| ZEV Biotech | yes | Plataforma de diagnóstico molecular modular para ser instalada en cualquier laboratorio | Desafíos para mantener y actualizar el software y firmware de forma remota en múltiples laboratorios |  |

### F · AI-native / computational — instrument: none (peer) — role: Peer: review, cite, co-author, endorse

| Company | AI/IoT base | What they do | Hypothesized gap (private) | Secondary fit |
|---|---|---|---|---|
| ArgenTag | yes | Tecnología de secuenciación de células individuales (single-cell) para diagnóstico e investigación | Falta de experiencia en ingeniería de software para crear plataformas SaaS accesibles para bioinformáticos | software-engineering gap |
| Beeflow | yes | Servicios y plataforma patentada de tecnologías y análisis de datos para optimizar la eficiencia de polinización | Dependencia de modelos predictivos no estandarizados y necesidad de escalar la plataforma de software SaaS |  |
| Biodiversity Intelligence | yes | Medición y documentación del impacto en la naturaleza combinando ADN y datos satelitales | Dificultad técnica para cruzar capas satelitales con reportes genéticos en una plataforma unificada |  |
| Bitgenia | yes | Plataforma de IA para el análisis e interpretación de secuencias de ADN con calidad clínica | Riesgo de estancamiento del algoritmo en prototipo sin una arquitectura cloud robusta y de alta disponibilidad |  |
| CellCo | yes | Desarrollo de medicamentos innovadores combinando simulación molecular e inteligencia artificial | Déficit en prácticas ágiles de ingeniería de software para transformar modelos académicos en producto comer… |  |
| CellRep | yes | Reprogramación celular basada en algoritmos de IA para terapias regenerativas | Dependencia de scripts de investigación locales en Python sin integración continua ni testing automatizado |  |
| Dogma Biotech | yes | Desarrollos terapéuticos basados en azúcares y moléculas descubiertas mediante IA | Baja madurez de gobernanza de datos y falta de pipelines automatizados para cribado virtual a gran escala |  |
| Elytron Biotech | yes | Plataforma biotecnológica de IA para acelerar el hallazgo de soluciones microbianas sostenibles | Falta de separación clara entre el stack de investigación bioinformática y el producto de software entregable |  |
| Geoprot | no | Análisis y predicción de la función de proteínas mediante un enfoque computacional sin IA | Software desarrollado por científicos con deuda técnica en rendimiento, modularidad y escalabilidad | software-engineering gap (scripts to product) |
| Mavios | yes | Digital twin impulsado por IA para la optimización y control de bioprocesos industriales | Dificultad para estandarizar la conexión con el hardware heterogéneo de plantas de fermentación |  |
| MIGMA | yes | Antioxidantes diseñados mediante IA para extender la vida útil de alimentos y cosméticos | Modelos predictivos aislados del proceso de formulación comercial y falta de gobernanza de software |  |
| Monte Caldera | no | Descubrimiento y síntesis de materiales avanzados mediante física computacional | Cuellos de botella en la ejecución de simulaciones físicas complejas por falta de orquestación cloud | software-engineering gap |
| Nanogrow | yes | Tratamientos biológicos con nanobodies integrando IA, modelado estructural computacional y biofísica | Foco excesivo en la bioinformática teórica con poca infraestructura cloud para despliegue y gobierno de mod… |  |
| Naturannova | yes | Plataforma guiada por IA para descubrir sabores y edulcorantes de origen vegetal | Enfoque orientado a laboratorio que descuida el empaquetado de la tecnología en un software escalable |  |
| OncoPrecision | yes | Plataforma para desarrollo de anticuerpos oncológicos analizando datos derivados directamente de pacientes | Cuellos de botella en la escalabilidad computacional y falta de estandarización de pipelines bioinformáticos |  |
| Qnity | yes | Sensores electroquímicos cuánticos para cribado acelerado de candidatos farmacológicos | Fuerte foco en el desarrollo físico cuántico descuidando la experiencia de usuario y la suite de software |  |
| Stämm | yes | Revolucionando la producción biológica y celular mediante fabricación continua, bioprocesadores y microfluídica | Riesgo de escalado de hardware sin plataformas robustas de observabilidad y sincronización de datos en tiem… | A (hardware partner) |
| Sylvarum | yes | Tecnología de electroestimulación en semillas para inducir vigor y mejorar la germinación | Hardware de laboratorio rudimentario que requiere profesionalización para operar en plantas de semillas | A (hardware partner) |
| Tesabio.ai | yes | Algoritmos de IA para traducir la biología de microARN en nuevas estrategias terapéuticas | Gran capacidad matemática/biológica pero falta de buenas prácticas de desarrollo de software y entrega cont… |  |
| Zavia Bio | yes | Plataforma computacional de diseño de proteínas para desarrollo de bioinsumos agrícolas | Sistemas algorítmicos desarrollados como scripts que carecen de empaquetado como producto comercial | software-engineering gap |

### Off-thesis — instrument: none — role: Ignore

| Company | AI/IoT base | What they do | Hypothesized gap (private) | Secondary fit |
|---|---|---|---|---|
| Cryosmetics | no | Cosmética natural personalizada con impacto positivo en el microbioma cutáneo | Falta de digitalización de la experiencia del cliente y trazabilidad rudimentaria de muestras |  |