Retrieval-augmented generation
Answers from aviation documents — with the page to prove it.
Ask about aircraft performance, weather minimums, regulations or procedures. Every answer is generated only from the indexed library and cites the exact document and page it came from.
System status
API server
FastAPI
Vector database
Supabase Postgres + pgvector
Embedding model
local ONNX
Language model
Groq
Recently added
View allHow answers are grounded
- 01Ingest
PDFs are parsed page by page; running headers and page numbers are stripped.
- 02Chunk
Section-aware chunks keep document, page range and heading metadata.
- 03Embed
A local ONNX model (bge-small) embeds chunks on CPU — no GPU needed.
- 04Retrieve
pgvector cosine search finds the most relevant passages for each question.
- 05Generate
Groq LLM answers using only those passages and must cite every claim.
If no passage is relevant enough, the system says the sources don't cover the question instead of guessing — and the language model is never asked.