05Internal AI knowledge assistant (RAG)·2025

Kauno Grūdai

An internal RAG assistant that answers from 1,000+ company documents — with exact source references.

OpenAIPineconeRAG1,000+ docsSource citationsHallucination safeguards
Problem

Employees across different departments were searching for answers in PDF and Word documents, so many repeated questions ended up with managers or responsible teams. They needed a fast, trustworthy search and answer tool grounded in internal sources.

What I did

I built an internal RAG assistant using OpenAI and Pinecone that lets employees query 1,000+ internal documents (PDF, Word) and get answers with links to the exact sources. I implemented strict source grounding with citation metadata and hallucination safeguards that route unanswerable questions to the relevant department instead of guessing.

Result

Employees find information faster across the 1,000+ document base, managers receive fewer repeated questions, and answers stay tied to clear sources.

Highlights
  • 1,000+ documents (PDF, Word) indexed, with strict source grounding — every answer links to its source.
  • Hallucination safeguards route unanswerable questions to the right department.
Internal company tool — shown descriptively (no public screenshots).