 helped leading technology companies test and monitor their LLM applications.

After failing to scale up GTM, we shutdown Gentrace and open sourced the codebase in Dec 2025.

A few highlights from my time working on Gentrace:

I guided customer (Walmart, Asana, Quizlet, Faire, Webflow, and others) engineering teams on AI evals and observability.

I architected an AI agent for debugging and rule creation, featuring a resource-based tool system with Cursor-like page context awareness, large trace payload traversal, and secure WASM sandboxed execution. I also added evals and observability (using Gentrace of course).

We built a self-hosted version using Docker and Kubernetes, culminating in a successful deployment into Walmart’s air-gapped environment.

I optimized CDC pipelines to ClickHouse for scalable span ingestion (>1k / s) and low-latency analytics reads (>100M rows).

I ceated a unified method abstraction for tRPC, agent tools, and REST APIs → SDK generation (via Stainless).