At IBM watsonx Orchestrate, I engineered hybrid lexical tool shortlisting in the Go agent runtime, raising top-1 tool selection from 63.9% to 80.6% with my evaluation harness.
I also traced agent routing failures to an interaction between skill gating and instructions, and diagnosed tool timeouts caused by a hidden execution-pool queue. My work includes designing agentic RAG workflows that improved answer accuracy by 28% in benchmarks.
Previously, I worked on payments at Revolut and Java/Spring Boot systems at Intellias, including microservices processing interbank transactions and a reporting backend for EV-charging analytics. I have also built a centralized log-analysis hub and a fault-tolerant Jenkins CI/CD pipeline.
