Director and systems architect. I build the software ecosystems that make custom silicon usable — compiler, runtime, kernels, and customer tooling — and run the engineering organizations that ship them. My career spans the full stack from ASIC tapeouts to agentic AI platforms.
At Flex Logix I directed the complete software stack for an AI inference accelerator: the Compiler, DevOps, and Solutions Architecture teams (20 engineers). Compiler infrastructure, operator kernels built against the hardware team's RTL macros, the runtime, the customer attach path, and the deployment and validation tooling customers used to get models running on our silicon. We grew model runtime support from 2 to 30 models. Before that I spent fourteen years in ASIC development at 3PAR/HPE — five silicon generations, three patents at the hardware/software boundary, zero respins on Gen6 — so I work with chip architects as a peer, not a customer.
At ZaiNar I built a multi-agent development platform. Specialized agents for code generation, dependency management, automated testing, and live system monitoring, working as a coordinated team alongside engineers. We rolled it across all five engineering teams.
I extended the same platform into a PM-facing agentic sandbox. Each product manager gets their own full-stack cloud sandbox cloned from the real product. They drive Claude Code to build features against a real API and database, deploy to a personal URL, and open a PR. Same agent platform, now serving PMs the same way it serves engineers.
Three years at ZaiNar guiding the org through a sequence of platform transitions. Rebuilt the software org into five functional teams. Re-architected the cloud platform from Lambda microservices to a NestJS modular monolith on ECS. Migrated the dev model to an agentic flow. Concentrated the team to the top ~40% of builders — 30 down to 12 — by eliminating contractor positions and roles the platform had automated away. Shipped the indoor location SaaS as a 3D real-time map experience: assets, equipment, and people moving on a floor model in real time, fed by an Apache Flink event pipeline. Most of the hard work has been on real-time data behaving correctly on a 3D scene.
Earlier I led a 30+ engineer software org at HPE through a multi-year platform re-architecture that shipped as HPE Primera in 2019, and a 12-engineer firmware/infra/DevOps team at AWS that replaced the legacy out-of-band server management network across the data center fleet. Both were major platform transitions. Customers depended on them. They shipped.
I'm currently building Tokenburner, a model-agnostic AI infrastructure platform on AWS Bedrock running at zero idle cost. Hands-on in the codebase, shipping alongside the team, owning the full system end-to-end.
tokenburner.md) ensuring AI assistants understand schema, conventions, and deployment target from first promptA decade of ASIC and RTL design before management. I'm not chasing silicon work today, but the hardware fluency still pulls weight — it's why I could lead compiler development at Flex Logix while engineers above and below the boundary trusted the calls. For teams building or integrating custom silicon, that lens still applies.