GenAI
Lexi-Chat — Document Intelligence
An enterprise document chat system with automatic fact extraction, tagging, and vector search over PDFs.
Problem
Teams accumulate PDFs faster than anyone can read them, and finding a specific fact buried in one means opening and skimming the whole file.
Solution
Built a Streamlit app that ingests PDFs via pdftotext, stores embeddings in Postgres with pgvector for similarity search, and uses GPT-4 to extract key facts, suggest tags, and answer natural-language questions with source references back to the original document.
Technology
- Python
- Streamlit
- PostgreSQL
- pgvector
- OpenAI GPT-4
- Peewee ORM
Result
Replaces manual document triage with searchable, taggable, chat-driven access to an entire document library.