AI search engine for a large global bank

Challenge

Employees at a global bank needed answers buried across huge, siloed internal knowledge bases. Classic keyword search returned long, unsorted hit lists and cost valuable time in day-to-day work.

Approach

Built a retrieval-augmented-generation architecture in Python with FastAPI, connected to a .NET service landscape and a React front end. Documents and vector metadata live in CosmosDB, running on Azure.

Outcome

Employees now get direct, source-backed answers instead of link lists. Result relevance and adoption rose noticeably, and the architecture keeps expanding to further internal search use cases.

// Tech stack

• FastAPI
• Python
• LLM
• OpenAPI
• .NET
• Azure
• React
• Ansible
• CosmosDB