I build AI systems that connect research with real-world use — from clinical RAG and agent engineering to distributed inference and full-stack applications.
- 🎓 CS @ UW–Madison, Class of 2027.
- 🔬 Leading technical development of a thyroid cancer AI support system at UW–Madison Surgery, focused on traceable generation and safety evaluation.
- ☁️ Previously built and evaluated tool-calling and multi-agent systems at Microsoft Cloud & AI.
- 🛠️ Currently exploring how everyday devices can work together to run larger models.
HeteroMesh — heterogeneous distributed inference
Connecting Windows, macOS, and iPhone to run larger models across heterogeneous hardware: distribute model weights, keep shards resident, and relay intermediate activations. Fully local video generation is the first application target.
Python · PyTorch · Swift · cross-platform systems
In development: foundational communication and real small-model relay are implemented; H3 partitioning passes small-configuration tests. The full video pipeline, H3 generation, and iPhone hardware inference are still to be validated.
Repository ↗ · System design ↗
| Project | What I'm building | Stack |
|---|---|---|
| Buzzer | A classroom party quiz platform with a shared host screen, mobile players, and live scoring. Team project. | FastAPI · Socket.IO · React · MySQL · Redis |
| Local Azure ReAct Agent | A local documentation agent with allowlisted tools, bounded recovery, source citations, and auditable traces. | Semantic Kernel · Phi-3 · Python |
| Financial Asset QA | Financial Q&A that routes questions to live market data or document retrieval, with sources and evaluation dashboards. | Next.js · FastAPI · FAISS · Tavily |
| Terruto | A Naruto Part I adventure mod with story progression, boss encounters, chakra mechanics, and pixel art. | C# · tModLoader |
More research and project stories on my website ↗
Also working with: LangGraph · Semantic Kernel · AutoGen · Milvus · BM25 · Kafka.
Guitar, cooking, baking, and basketball. I like building things people can use — and occasionally worlds they can get lost in.
Let's talk about AI applications, agent systems, research, or an interesting idea worth building.