A natural-language analytics assistant on top of lakehouse data, orchestrated with LangGraph and Databricks Model Serving and grounded by Qdrant on Azure Container Apps.
LangGraph
Agent orchestration
Databricks
Model serving + Unity Catalog
Qdrant
Vector store on Azure
MCP
SQL + RAG tools
A financial-data analytics provider needed a production AI assistant that lets analysts ask questions in natural language across governed lakehouse data. We built a LangGraph-based agent that calls Databricks Model Serving for LLM inference, Qdrant on Azure Container Apps for vector retrieval and a set of SQL/RAG tools exposed through MCP, all wired to the existing identity and observability backbone.
Client
Financial-data analytics provider
Industry
AI & Data Platforms
Scope
RAG agent + vector store + MCP tools + serving integration
Stack focus
LangGraph, Databricks Model Serving, Qdrant, MCP, Python, TypeScript
Putting AI on top of governed banking data demands strict identity, traceable answers and a tight integration with the existing lakehouse.
Analysts must trust each answer with a clear citation back to gold tables and documents.
Model serving has to be reliable, monitored and aligned with banking-grade identity.
The assistant needs SQL, RAG and structured tools — not just a chat completion endpoint.
We delivered a production architecture with the right primitives for governed AI on lakehouse data.
Stateful agent that orchestrates retrieval, tool calls and response shaping.
Managed model serving on Databricks for low-latency, governed inference.
Vector store for RAG retrieval with autoscaling on Azure Container Apps.
Tooling layer exposed via MCP so the agent can run governed SQL and document retrieval.
Tied into the bridge platform's Auth0 + Azure identity for end-to-end auditability.
The assistant runs against governed lakehouse data and gives analysts a trustworthy natural-language interface to their domain.
Grounded
RAG over governed data
Production
Lakehouse model serving
MCP-native
Reusable SQL + RAG tools
Identity-aware
Tied to platform SSO