System Software Engineer at HP · Edge AI & Agentic Systems
I build AI systems for real-world work. At HP, I work on bringing agentic capabilities to PCs through local model serving, NPU-backed inference, and performance evaluation. I built a benchmarking pipeline and dataset to compare knowledge-worker agent workloads across models, chipsets, and agent harnesses.
Outside work, I build open-source products and developer tools that make AI more useful—and easier to work with.
| Project | What it does |
|---|---|
| Roomfile | An interior-design agent skill that remembers a room, develops concepts, checks measured furniture fit, and turns ideas into a sourcing plan. |
| HelmMate | A local console for AI coding work: agent-ready tickets, dependency checks, isolated worktrees, run tracking, and human review. |
| Cairn | Project-specific memory for coding agents, shaped through a repository interview and maintained through reviewed session updates. Early-stage, actively dogfooded. |
| Local RAG | A local-only document Q&A demo with Ollama, retrieval, conversation context budgeting, and inference performance instrumentation. |
| MINT | A causal-intervention framework for tracing where visual and textual information fuse inside multimodal language models. |
| Banana Codex | A Codex-native image-generation skill with Gemini integration, setup validation, and cost tracking. An unofficial derivative of Banana Claude, with upstream credit. |
- On-device AI: local inference, model serving, NPU execution, and system resource tradeoffs.
- Agent systems: useful workflows, clear execution boundaries, persistent context, and human review.
- Evaluation: reproducible workloads and experiments that inform engineering decisions.
My background includes multimodal AI research at Rice University’s Chili Lab, enterprise AI engineering at Systex, and cloud and MLOps work at Microsoft.
Always happy to exchange ideas about practical AI and open-source tools.