Retrieval-Augmented Generation on Azure and Databricks

The challenge

The client wanted to test whether generative AI could provide reliable answers grounded in its own data.

The approach

Kumulus prepared the data sources in Databricks and built a retrieval-augmented generation solution on Microsoft Azure, with access controls and validation of responses.

The outcome

The team can now test semantic search and grounded answers over the organisation’s own content, giving a practical foundation for a future production assistant. A working RAG flow integrates Azure AI Search and OpenAI with Databricks and the existing data lake, covering ingestion, vectorisation and query.