Balancing AI performance and accountability in financial services

How can we strike the right balance between rapid AI innovation and responsible oversight in financial services?

Speaking at Deutsche Bank's Innovation Day 2025 in Singapore, Matteo Pelati, Co-founder of LangDB.ai, identified key areas for balancing AI performance and accountability in financial services : establishing guardrails for agentic AI applications, implementing real-time evaluation to improve accuracy and agent behavior, and ensuring effective auditing to refine models and boost application performance.

Here's the full transcript of the video:

“There are few important aspects actually that came out of the discussion.

One is guard railing your AI system, so putting the proper control in place to make sure your agentic application is behaving correctly. And this is becoming more and more important now that you're not simply limiting your interaction with the LLM, but it's a full agentic application which can take a complex decision and reasoning.

The second aspect is evaluation. You want to be able to evaluate if the response that has been given are right or wrong. And you might want to be able to do this in real time to force a particular behavior or prospects to analyze and further improve your agent.

And the third aspect is auditing, not simply for the sake of going back and check what happened, but also because all this data can actually be used to fine tune models and improve your entire application."