Full-stack reference build
StatmateAI
A full-stack AI assistant that automates statistical analysis for clinical and observational research — LLM agents that select tests, run them, and explain the results.
- FastAPI
- LangGraph
- PydanticAI
- SQLAlchemy
- Streamlit
- Docker
StatmateAI combines LLM agents with classical statistical methods so researchers can upload data, run rigorous analyses, and interpret results without hand-wiring every test. It's also a deliberate demonstration of production architecture end to end.
What it does
- Automated test selection — agents inspect the data and recommend the appropriate statistical test, then run it.
- Interactive UI + REST API — a Streamlit front end over a FastAPI backend with 18 endpoints for programmatic access.
- Comprehensive results — statistical trees, p-values, effect sizes, and natural-language summaries that a non-statistician can act on.
- Scheduled jobs — background and recurring analyses via APScheduler.
Built for
Clinical researchers with small-sample studies, observational research needing rigorous statistics, and teams that want reproducible, auditable analysis workflows.
Why it matters
It's a reference for how I build GenAI systems for regulated, high-stakes domains: typed agent boundaries (PydanticAI), an explicit workflow graph (LangGraph), SQLAlchemy storage (SQLite/Postgres-ready), Docker Compose, GitHub Actions, and Prometheus/Grafana monitoring — the production scaffolding, not just a prompt.