Atlagene
Atlagene is a genomic health analysis platform. People upload their DNA file (23andMe, Ancestry, whole-genome), and the pipeline returns a physician-reviewed report covering 21 health categories. Founded 2026, U.S.-based.
The problem
Millions of people sit on raw DNA files they paid for and have no way to read them. Generic health advice ignores what's in someone's genome. Specialist genetic counseling is expensive, slow, and not built for the consumer use case.
Existing direct-to-consumer tools split into two camps: ancestry maps (interesting, not actionable) and narrow disease panels at clinical price points. Nobody connects raw consumer data to a useful, defensible daily-life picture.
Approach
Atlagene is a polyglot system. The web app handles upload and the report UI. A TypeScript Fastify API runs auth, billing, and orchestration. A Python pipeline does the actual genomics: BioPython and pysam to parse the file, a curated variant registry to score it, scikit-learn and XGBoost models to weight the results, then an Anthropic-assisted layer to compose the human-readable explanation. A clinician reviews each report before it ships.
The infra is built around cost. The API runs on small always-on EC2; the pipeline runs on Fargate as pay-per-job, so analysis cost scales with revenue rather than with peak capacity. Privacy is first-class: genomic data is encrypted end-to-end and isolated from the user-facing services.
- Next.js web app with secure DNA file upload
- TypeScript + Fastify API (auth, billing, orchestration)
- Python + FastAPI genomics pipeline on AWS Fargate
- BioPython + pysam for variant parsing across consumer formats
- scikit-learn + XGBoost models scoring across 21 categories
- Anthropic-assisted report composition, physician-reviewed
- PostgreSQL primary + Redis queue + S3 cold storage on AWS
Status
Atlagene is live at atlagene.com and onboarding users. As founder and CEO I lead product, engineering, and the clinical-advisory partnerships that keep the report defensible.
What I'm learning
Healthtech is half product, half compliance, half clinical credibility. Yes, that's three halves. The trust budget you spend on a single misleading insight is enormous, so the bar for shipping anything is much higher than in pure SaaS.
A polyglot stack is a tax you pay for honesty. You can't do real genomics in TypeScript, and you can't do a great web product in Python. Splitting them was the right call, even though it doubled the deploy story.
Founder-CEO is a different job from founder-engineer. The lever shifts from writing code to choosing what doesn't ship.