← BACK TO PROJECTS
Onpoint Insights

AI Document Assistant: Search Across 20+ Sources

Client teams couldn't answer questions across 20+ scattered documents with standard keyword search.

17% more accurate than standard keyword search
HOW THE SYSTEM WORKS
IN
Team question
about any of 20+ documents
STEP
Intent router
picks which documents matter
DATA
FAISS vector index
documents as searchable embeddings
STEP
Grounded LLM
answers from retrieved passages only
OUT
Answer with sources
17% better than keyword search
Questions are routed to the right documents before the AI answers, so answers stay grounded.
Built
AI assistant that retrieves answers across documents, with intent routing to the right source.
Approach
Transformer embeddings + FAISS vector index + LLM-driven intent routing.
Impact
17% more accurate than standard keyword search
Skills
FAISSTransformersRAGIntent Routing