Work
Kastle
I’m a software engineer at Kastle, where I build AI agents and the software used to configure, run, and evaluate them. My work spans frontend interfaces, backend services, and model infrastructure.
I work on tools for defining agent behavior: visual workflow editors, agent configuration, and forms and extraction rules for capturing information from conversations. I build the underlying data models and APIs alongside the interfaces, connecting what someone configures in the product to what happens when an agent runs.
I also work on evaluation and debugging. That includes datasets and annotation tools, checks for unsupported claims and missing information, and ways to inspect a response against the evidence available to the model. Activity logs and analytics help make agent behavior easier to understand and improve.
On the infrastructure side, I’ve worked on model routing, fallbacks, streaming, and observability. This supports comparing models, diagnosing failures, and evaluating changes before rolling them out.
My work also includes live assistance for human agents, connecting streaming audio and telephony systems to information extraction, summaries, and the agent-facing interface.
Projects
Kubrick — semantic search for video.