Skip to content

InteLIS Insights

Natural-language analytics for laboratory data. Free, open source, self-hostable.

Ask questions about lab results in plain English; get answers as charts and tables. Built for ministries of health, donor programs, and public-health labs that already run InteLIS and want AI on top of their existing data — without paying for SaaS, without sending patient data to a vendor, and without their data leaving their infrastructure.

Licence

InteLIS Insights is licensed under AGPLv3. Source code, audit, modify, self-host — all free. See LICENSE on GitHub.

What it does

You point it at your existing InteLIS MySQL database (read-only). Then:

  1. A user signs in and asks a question — "What's the VL suppression rate by district last quarter?"
  2. The app classifies intent, retrieves relevant schema context from a vector DB, and generates SQL using your chosen LLM (OpenAI, Anthropic, Google, Mistral, DeepSeek, Groq, or a local Ollama model).
  3. The SQL goes through access-control checks (a district-level user can't query province-level data) and safety validation (SELECT-only, no PII columns, no patient identifiers).
  4. The validated SQL runs against your lab DB. Results come back as a table plus an auto-suggested chart.
  5. Every step is logged for audit.

Privacy by architecture

No patient identifiers are ever sent to the LLM. The LLM never holds a database connection. Only the user's question and the database schema are transmitted; results stay in your infrastructure.

Built on

Industry-standard, permissively-licensed FOSS the whole way down. Any full-stack TypeScript engineer can read and extend this codebase on day one — no proprietary runtime, no SaaS dependency, no obscure framework.

Next.js 16 · LangGraph.js · Vercel AI SDK · Qdrant · PostgreSQL + Drizzle · Auth.js v5 · shadcn/ui + Tailwind CSS · Recharts · LangFuse · Node 22 LTS

See the full stack and the rationale on the Architecture page.

Contents

  • Getting started — Docker Compose, an existing InteLIS database, one provider key.
  • Configuration — Every environment variable, what it does, and which provider needs what.
  • Architecture — The stack, the components, and how they fit together.
  • Query flow — How a question becomes a chart, step by step.
  • Codebase tour — Folder-by-folder walkthrough for contributors new to the stack.
  • Privacy & RBAC — How patient data stays safe; how district / province / national tiers are enforced.
  • LLM providers — Supported providers (OpenAI, Anthropic, Google, Mistral, DeepSeek, Groq, Ollama) and how to pick.
  • Contributing — All TypeScript / JavaScript. Built to be easy to read, easy to extend.
  • Implementation plan — Current progress, known gaps, v2 roadmap.