I turn coffee
into science.

The platform I lead has served over 1.3 billion requests and grown 55× year over year in adoption, on a GPU fleet across three datacenters we scaled roughly 6×. We run open-weight models up to 3 trillion parameters with 1M-token context on our own hardware, which is what keeps sensitive workloads off third-party infrastructure. Underneath it sits a gateway that unifies several commercial LLM providers and our on-premises serving behind a single API and token, with per-token concurrency and spend governance.

I stay close to the work. I founded our document-intelligence product and built the analyzer service at its core: a Kafka-driven pipeline that preprocesses and OCRs documents, routes them through vision and reasoning models for structured extraction, and validates the results. I also own the LLM safety and behavior layer across our assistants: topic rails built from real support corpora, off-topic detection, multilingual refusal handling. That layer is what turns a demo into something legal and security will approve.

Before this I spent my career in applied R&D. At Tesla I led Vision R&D, building automated inspection and defect detection deployed across the Giga Factories. At Honeywell I led teams that shipped AI products to market in gas leak detection, thermal imaging, warehouse robotics, and industrial IoT.

My Ph.D. is in physics from Ohio University, studying nuclear outbursts in the centers of galaxies: extreme variability in the supermassive black holes powering active galactic nuclei, including a Seyfert galaxy caught in the act of rapidly transitioning from one type to another. That training is why I default to measurement: define the metric, run the experiment, let the data decide.