Back to Projects
Agentic AIadvanced

PolicyPilot : Agentic RAG with LangGraph on Azure Databricks using Unity Catalog, Vector Search, Unity AI Gateway and Databricks Apps

PolicyPilot is a governed, agentic Q&A copilot that turns dense regulatory filings into answers you can actually use. Ask it about any of 20 large companies — for example, “What supply chain risks does Apple disclose?” — and PolicyPilot searches the company’s latest SEC 10-K, builds an answer from the filing, and attaches a numbered citation to every claim. If the filing doesn’t support the answer, it doesn’t improvise. PolicyPilot refuses instead of guessing. Under the hood, it runs as a LangGraph agent on Azure Databricks. Unity Catalog governs the data, Vector Search handles retrieval, Unity Gateway sits in front of the LLM, and Databricks Apps serves the copilot to users. GitHub Actions handles deployment with no stored cloud credentials. The result is a regulatory Q&A system designed to be grounded, traceable, governed, and deployable in production.

19 lectures

What You Will Learn

Design an agentic RAG workflow in LangGraph with plan, retrieve, answer and verify steps
Build citation gates that enforce grounded answers and refuse unsupported claims instead of hallucinating
Govern data and AI assets with Unity Catalog catalogs, schemas, tables and permissions
Build a Delta Sync Vector Search index using self-managed embeddings
Route LLM requests through Unity AI Gateway without exposing provider API keys to the application
Deploy Databricks apps and gateway resources as code using Databricks Asset Bundles
Build keyless CI/CD with GitHub Actions OIDC, Azure and Key Vault
Diagnose real deployment issues involving Azure RBAC, service principals and Unity Catalog permissions

System Architecture

PolicyPilot : Agentic RAG with LangGraph on Azure Databricks using Unity Catalog, Vector Search, Unity AI Gateway and Databricks Apps Architecture Diagram

High-level architecture overview of the PolicyPilot : Agentic RAG with LangGraph on Azure Databricks using Unity Catalog, Vector Search, Unity AI Gateway and Databricks Apps .

What You'll Build

  • PolicyPilot, a governed Q&A copilot that answers questions from SEC 10-K filings with citations
  • An ingestion pipeline that fetches, chunks and embeds real SEC 10-K filings for 20 companies
  • A LangGraph agent that understands company names from natural language and retrieves from the correct filing
  • A local setup using ChromaDB and Groq that switches to Databricks with a single environment variable
  • A Unity Catalog Delta table and Vector Search index for production-grade retrieval
  • A Unity AI Gateway model service for Groq's gpt-oss-120b, fully deployed from YAML
  • A Streamlit chat application with citation sources, deployed as a Databricks App
  • A CI/CD pipeline with automated tests, Databricks Agent Evaluation and one-click OIDC deployment

Project Instructor

Yash Patil

Yash Patil

2+ years exp
LinkedIn
PolicyPilot : Agentic RAG with LangGraph on Azure Databricks using Unity Catalog, Vector Search, Unity AI Gateway and Databricks Apps
Premium
One Subscription. 40+ Projects. Unlimited Access.
AccessMobile & Web