Building Self-Improving Agents
Most agent failures go unreported: slow replies, doom loops, bad tool calls, hallucinations. The few that do get reported are fixed by hand, one prompt tweak at a time.
In 45 minutes, we will build a support agent that improves its own prompt, live. We'll cover
A durable agent loop on Temporal workflows
Failure signals from OpenTelemetry traces in Oodle
New prompt versions drafted from Oodle insights
Full-agent evals on a golden dataset, scored by code
Slack approval through Temporal signals before production
About the speaker:
Gaurav Maheshwari is a Founding Engineer & India Site Lead at Oodle AI - an Observability startup building a new data layer for Observability for Agentic Era. He has previously worked at Rubrik and Goldman Sachs. Beyond work, poetry and cricket interest him.