Retrieval
RAG & Knowledge Systems
Retrieval quality decides answer quality. Embedding everything and hoping cosine similarity sorts it out produces a demo, not a system. We build hybrid retrieval that combines semantic and keyword search, chunking that respects document structure, and output that cites its sources.
- pgvector
- PostgreSQL FTS
- Docling
- PyMuPDF
- FastAPI
What this includes
Each of these exists because leaving it out is how these systems fail in production.
Hybrid retrieval
Vector search over pgvector with HNSW indexing, fused with PostgreSQL full-text search via Reciprocal Rank Fusion. Semantic search finds meaning; keyword search finds the exact part number. You need both.
Structure-aware parsing
Documents parsed with Docling and a PyMuPDF fallback, then chunked along section and table boundaries rather than every N characters — so a table never gets cut in half.
Citations that hold up
Every answer carries page numbers and section headings back to source. When someone challenges an answer, you can show exactly where it came from.
Live ingestion
Agents that pull in a URL or a document mid-conversation and can reason over it immediately, with presigned direct-to-storage uploads for large files.
RRF
Vector and full-text fused, not chosen between
HNSW
Indexed pgvector search
512
Max tokens per structure-aware chunk
Next step
Tell us what you’re trying to build
Most engagements start with a fixed-price audit, so the first thing you buy is a decision rather than a commitment.