Internal Knowledge Assistant
LLM Integration · 3hrs saved per employee per week
Employee onboarding time reduced by 40%, and senior staff reported significantly fewer interruptions as the assistant handled routine knowledge queries.
The challenge
A professional services firm with 200+ employees had institutional knowledge trapped in thousands of documents across SharePoint, Confluence, and email chains. Finding accurate answers to policy, procedure, or client-related questions took employees an average of 25 minutes per query.
New hires took 3-4 months to become productive because knowledge transfer was informal and inconsistent. Senior staff were constantly interrupted with questions they'd answered dozens of times before, creating a productivity drain across the entire organisation.
Our approach
Built a retrieval-augmented generation (RAG) system that ingests and indexes the firm's entire document corpus using vector embeddings stored in Pinecone. The system processes PDFs, Word docs, wiki pages, and email archives.
Integrated a fine-tuned LLM that generates precise, cited answers — every response includes source document references so employees can verify and dive deeper. Answers are scoped to the user's access permissions.
Deployed as a Slack bot and web interface, meeting employees where they already work. Usage analytics track which knowledge gaps are most common, informing documentation improvements.
Results
3hrs — saved per employee per week
89% — answer accuracy rate
45sec — average query response time
Stack
OpenAI API, Python, Pinecone, FastAPI