AI Projects

Policy Assistant with Retrieval and Human Approval

A

AI Builders Team

Community Starter · Jun 10, 2026

Goal: Help employees find and apply company policies with grounded citations and human sign-off. Plan: - Data: HR and Security policies in Confluence/SharePoint; nightly sync to object store. - Pipeline: Clean, chunk (section-aware), embed, index; add effective dates and owners in metadata. - App: Chat UI with suggested queries; answers must include section and link. If high-risk topic, require approver review before sharing. - Stack: Python FastAPI, Weaviate or Qdrant, bge embeddings, reranker, GPT-4o or Claude 3.5 for generation. - Governance: Redact PII, log citations, store feedback. Version prompts and restrict tools by role. - KPIs: Time-to-answer, deflection rate from HR tickets, hallucination rate, user trust scores. Roadmap: Add policy change diffs, proactive alerts, Slack integration.

💬 0 replies👁 497 views

💬 0 Comments

Login to join the conversation

Login to Comment
No comments yet. Be the first to share your thoughts!

Related Discussions