GINA
Grounded insight from natural-language analytics
Ask a question in English, get a real query — and the grounds it was built on.
What it does
- Built the backend and query orchestration turning natural-language questions into executable SQL and grounded insight.
- Implemented SQL safety before anything reached the database: AST validation, SELECT-only enforcement and table whitelisting.
- Grounded generation in the uploaded dataset’s own schema using schema profiling, embeddings and pgvector retrieval.
- Implemented Server-Sent Events for streaming query progress and responses.
- Worked on response caching, LLM fallback strategies, telemetry and read-only PostgreSQL execution.
- Integrated the authentication and authorisation checks around conversations and data access.