Birbal
An agent harness with model adapters, tool execution, configurable pipelines, evaluations, and tracing. It also powers Birbal Brief, my AI research newsletter.
I build AI systems, products, and teams.
Most of it sits somewhere between hands-on engineering, technical strategy, and building teams.
A few projects I'm building now: an agent harness, on-device photo search, and a local codebase briefing tool.
An agent harness with model adapters, tool execution, configurable pipelines, evaluations, and tracing. It also powers Birbal Brief, my AI research newsletter.
An iOS app for on-device semantic photo search. I built the Core ML conversion, local inference, OCR, and time and location search.
A local desktop app that creates role-specific codebase briefings with source citations. It indexes repositories, builds a memory graph, and works with local LLMs.
A few roles across startups, enterprise software, and open-source infrastructure.
At Proferent, I work with companies on the practical side of AI adoption: choosing useful problems, redesigning workflows, and getting systems into production.
Visit ProferentBuilt and led a 30+ person global solution engineering organization supporting 50+ blockchain startups through product architecture, technical delivery, and production launch.
Built hyper-scale enterprise IoT systems processing 1M+ messages per second and advised customers on cloud migration, application modernization, and solution architecture.
Recent writing on AI adoption, AI systems, and engineering.
Sonar reads any codebase and writes a source-grounded briefing for your role. Every claim is cited to a file and line, and it runs locally on your machine.
We created fear around AI replacing jobs, then incentivized token usage over outcomes. The result? Everyone is showing off AI workflows, but nobody is talking about what …
Enterprise AI has a productivity story that is being advertised quite a bit, but it also has a cost story. Let's dig into that.