One-Page Grant Proposal · 2026

A Biotech-Grade AI Reasoning Platform for Regulated Life Sciences

Johnny Efrain Ortega

Founder & Sole Engineer · Age 15

Independent Research Project

Funding Sought: $50,000 – $275,000

Applicant

Johnny Efrain Ortega, 15 years old — self-taught full-stack engineer, sole author of the platform and its 14 verification cycles.

Project

A verifiable, audit-ready AI system for biostatistics, dose modeling, assay validation, CMC, pharmacovigilance, and 10+ other regulated biotech workflows.

Stage

Working prototype. 12 solver modules live, 1,835 regulatory chunks indexed, deterministic math verifiers passing golden tests, HMAC-signed audit exports.

Ask

Non-dilutive grant funding to reach validated design-partner pilots with 2 biotech firms within 12 months.

Problem

Biotech R&D runs on statisticians, regulatory writers, and QA reviewers who spend 40–60% of their time on repetitive artifact production: Statistical Analysis Plans, dose-response memos, method validation reports, deviation write-ups. General-purpose AI (ChatGPT, Claude) is fluent but unverifiable — it hallucinates citations, drifts on math, leaves no audit trail, and cannot be signed off by a QA lead. Regulated companies therefore cannot use it for anything that touches a submission.

Solution

I built a full-stack platform that pairs LLMs with deterministic verifiers and a locked regulatory corpus, so every answer is grounded, math-checked, and cryptographically auditable. The system refuses to answer when citations aren't in the corpus or when its own math drifts >2% from a native TypeScript stats engine (validated against R/SAS golden values).

What Is Already Built (Solo, in ~6 months)

12 solver modules: Power/Sample Size, PK/Dose, Assay Validation, Stability, Process Validation, Method Validation, Comparability, CMC, Pharmacovigilance, QMS, Freeform Chat, plus a batch runner.
Deterministic numeric verifiers covering non-inferiority, survival, Bayesian assurance, AUC, and 10+ other biostats domains — parity-tested against textbook R/SAS output.
1,835-chunk pgvector corpus of 21 CFR + curated regulatory sources with section-level citation guarding.
Compliance posture: PHI/PII firewall, immutable audit trail with DB triggers, HMAC hash-chained export bundles, two-person reviewer workflow, pinned prompt/verifier version snapshots.
Multi-tenancy: orgs, roles, Row-Level Security across every table.
Infrastructure: SSE token streaming, multi-model fallback + circuit breaker, durable L2 cache, telemetry, public REST API with Bearer keys and monthly quotas, self-host Docker + Helm packaging with a pluggable model adapter (OpenAI / Azure / vLLM).

Why Me

I am 15 years old. I taught myself TypeScript, Postgres, pgvector, cryptographic audit design, and biostatistics well enough to reproduce R output in native code. I designed and shipped the entire system — schema, RLS, streaming pipeline, verifiers, corpus ingestion, admin dashboards, and self-host packaging — without a team, a co-founder, or outside funding. This proposal exists because the technical work is already done; what remains is not code but validation, licensing, and expert sign-off.

Use of Funds (12 months)

• ICH / CLSI / USP corpus licensing & ingestion

$40k

• Golden R/SAS eval set (biostatistician-authored, ~500 cases)

$45k

• QA lead: IQ/OQ/PQ validation package + SOPs

$55k

• SOC 2 Type I readiness + HIPAA BAA counsel

$35k

• Biostatistician retainer for prompt/verifier sign-off

$40k

• Infrastructure, security tooling, incorporation, misc.

$20k

• Contingency & runway for 2 design-partner pilots

$40k

Total requested$275,000

12-Month Milestones

Q1 — Delaware C-corp, SAM.gov, licensed ICH/CLSI/USP ingestion complete.
Q2 — Biostatistician-signed prompt set; 500-case golden eval >98% accuracy in CI.
Q3 — Signed QA validation package (IQ/OQ/PQ) + SOC 2 Type I readiness report.
Q4 — Two biotech design-partner pilots on self-hosted deployments; letters of intent.

Impact

A verifiable AI layer for regulated biotech reduces statistical-analysis-plan authoring from weeks to hours without introducing submission risk. Beyond commercial value, this project demonstrates that a self-taught teenager, given non-dilutive support, can build serious regulated-industry infrastructure — a model for early-career American technical talent.