RAG Knowledge Assistant
Internal chatbot answering policy, HR, and engineering questions over 50k+ documents with citations.
The Challenge
Employees couldn't find answers across SharePoint, Confluence, and ServiceNow, generating thousands of repeat L1 tickets every month.
Our Solution
Ingestion pipeline that chunks and embeds 50k+ documents into pgvector, with a Claude-powered RAG chat that always cites its sources and respects per-document ACLs.
Key Features
- Source-cited responses
- Per-document ACL enforcement
- SharePoint/Confluence/ServiceNow ingest
- pgvector hybrid search
- Slack + web chat surfaces
- Feedback-driven re-ranking
Our Process
- 1
Ingestion
Built connectors with ACL preservation.
- 2
Retrieval
Tuned chunking, hybrid search, and re-ranking.
- 3
Answering
Claude with strict citation prompts and refusal logic.
- 4
Adoption
Rolled out via Slack first to drive habits.
Results
- L1 ticket volume down ~40%
- Average answer time: under 5 seconds
- Every response includes citations users can verify
"It's the first internal tool we've shipped where employees actually trust the answers."
Related Projects
Agentic AI Recruiting Copilot
Multi-agent system that screens resumes, ranks candidates against JDs, and drafts personalized outreach — with human-in-the-loop review.
Invoice & AP Automation
AI pipeline that ingests vendor invoices via email, extracts line items, validates against POs, and posts to ERP.
Sales Outreach Agent
Autonomous agent researches accounts, drafts hyper-personalized email sequences, and books meetings into the CRM.
Want a project like this?
Tell us what you're building. We'll show you how we'd approach it.
Start a Conversation