The Sufferings Before ExoBench, and the MCP it takes to Solve Them

One hectic Monday morning I asked an AI to speed up a dashboard query while buried in more urgent sprint tasks, staging got twice as fast so I nearly shipped it. At the last second I asked a DBA to log in and double-check. The 8 seconds became around 150 and I barely pulled the plug. A few weeks later a close cousin of my dashboard query made its way into our Hibernate stack. The entities were clean, the tests were green, and the ORM quietly emitted one SELECT per row which none of the testers quite bothered to worry about. I would have shipped that code too.

In this talk I want to talk about what it took to build the MCP product ExoBench that solves these kinds of issues. First I am going to talk about MCP techniques that it takes to build this kind of product. How to craft tool-descriptions whose length needs to far exceed anything that Claude or ChatGPT will actually read into their context windows using compact examples and documentation-endpoints. How to get AI agents to actually give you correlation-ids and call your endpoints without timing-out if they need to run for more than a minute (or 30 seconds or however much Anthropic/OpenAI have the patience for). I’ll also talk about how to effectively version MCP endpoints and how to use server-instructions, all in the context of Spring AI.

Then I’ll talk about how ExoBench works, what kinds of problems it solves, and it can help you optimize queries and reduce N+1 issues effectively. I’ll go into the darker topics of SQL benchmarking and data-skew. Finally, you’ll learn about how to use ExoBench to plunge into your codebase and stop N+1 issues before they happen. Have any nasty SQL?

Bring the query you'd ship tomorrow.

About the Speaker

Alexander Ioffe acquired his formative scars in the Java EE and big-data trenches, wrestling with MapReduce search indexes at eBay, object-relational impedance mismatches, and sprawling SQL in Nasdaq’s SEC-compliance systems. Those battles convinced him of the merits of functional programming, immutability, and composition.
Hungry for cleaner abstractions, Alexander plunged into the Scala ecosystem, embracing compile-time metaprogramming and LINQ-style query builders, and ultimately found his way to Kotlin. Drawing on years of meta-programming experience, he created ExoQuery, a library that brings true LINQ-grade, type-safe queries to Kotlin—with capabilities that had heretofore existed in no language or platform.

About the Sponsor

Certara (https://www.certara.com)
Certara accelerates medicines to patients, using biosimulation (modeling and simulation) and technology-enabled services to transform drug discovery and development. Our integrated and proprietary end-to-end platform with generative AI technology, biosimulation, regulatory science, and market access solutions combined with our global 1,400+ strong team of scientists and experts enable informed decision-making, higher R&D productivity, and improved patient outcomes.