I'm a Senior Technical Staff Member at IBM Research in Austin, TX. I build AI-native systems: agent harnesses, MCP-connected platforms, and knowledge graphs that turn large, messy domain data into something models can learn from.
What I'm working on
- AI for chip design: curating RTL, verification, and physical-design data into knowledge graphs that reinforcement learning teams use to train models.
- Agent guardrails: turnstile enforces a planner → reviewer → executor → test-runner pipeline in OpenCode through runtime hooks rather than prompts. Policy steers; turnstile blocks.
- Local inference (local-agent-stack): I run open-weight LLMs on a Mac Studio M3 Ultra (512 GB) with oMLX. Current rotation: GLM-5.3-Flash (oQ4) and Qwen3.8-Flash-Next (oQ4), Ornith-1.5-35B-A3B and Qwen3.8-27B (oQ8), all with MTP speculative decoding, at roughly 35–100+ tok/s depending on the model. I orchestrate agents across parallel worktrees with Orca.
Before this
- Built EDAaaS, an AI-native hybrid-cloud platform for chip design: agent-generated Terraform, an MCP server for workflow authoring, and root-cause agents.
- Architected the IBM Functional Genomics Platform, with ~400K genomes, 70M+ genes, and ~300M protein domains.
- Built Watson solutions such as Chef Watson and the Watson Advisor for Leukemia. Along the way: 16 granted patents and 20 peer-reviewed publications.
Tools I use daily: Python · TypeScript · Go · Rust · Claude Code · OpenCode · IBM Bob · MCP · Terraform · OpenShift · IBM LSF