Building an OpenSearch Research Agent at Scale

A lot of teams are now working on the same (unsolved) problem: exposing to an agent all the data of their company. This is essential to automate workflows, surface information, and become an AI-Native company.

I've been working on this problem for almost a year with a small team of senior engineers, and in this talk I'll discuss the architecture we built to achieve this goal finally.

We'll cover everything: from the design choices that allowed us to give a real semantic meaning to OpenSearch fields, how we collect and leverage unwritten knowledge by SMEs, how we evaluate the system, and how we keep it safe with a reliable permission-management system.

What started as a client project became an optimized harness to work with very big OpenSearch clusters - and in this talk, I share all the technical decisions that led to a production system you can trust!