Data Product Lead - AI Document Search
My role and scope
I was responsible for the internal document-search product, from studying consultants’ needs to developing the application and training users.
Context
The firm wanted to query a document collection of over 150 GB while keeping data processing local. The product combined document retrieval with a language model, using a RAG approach.
Product design
- Analyzed the document collection and consultants’ search needs.
- Defined local-processing constraints and session workspaces.
Development and user training
- Built the ingestion pipeline, including file conversion, text chunking and local vector storage.
- Developed the FastAPI backend and React interface.
- Integrated a local language model on dedicated GPU infrastructure and adjusted vector retrieval.
- Prepared training materials and trained consultants to use the application.
- Defined checks for ingested documents and monitoring of the vector database.
Application and training materials
- Document-search application with local processing and session workspaces.
- Ingestion and vector-search pipeline with document quality checks.
- Training materials for consultants.