Kira.

The context

Strategic decisions in a large insurer require locating, consolidating, and interpreting information from different sources. Kira was designed to support that process in a complex, regulated business environment, with attention to the quality and governance of the information used.

My contribution

I worked as a Software & AI Engineer on the development of the solution. My contribution included building an MCP Server to provide controlled access to tools and data within a multi-agent architecture.

I also worked on evaluation using Azure AI Foundry, following groundedness, fluency, similarity, and coherence metrics to assess the solution’s responses.

Engineering approach

The solution combines RAG and access to structured and unstructured data. Text-to-SQL agents support interaction with structured sources, while the MCP Server provides an interface to tools and data.

This brings data access and response evaluation into the same engineering effort: making information available to the agents and assessing the quality of what they produce.

What it enabled

Kira was designed to reduce the operational effort involved in finding and interpreting critical information, and to improve the consistency and confidence of decision support.

The solution won Avanade’s FY26 Q3 Hall of Fame — Global Solutioning and Delivery, recognizing the work of the team.