Hierarchical Spatio-temporal Visual Analysis of Cluster Evolution in Electrocorticography Data
Best PaperACM-BCB · ACM Conference on Bioinformatics, Computational Biology & Health Informatics · 2016
ML / AI Research Engineer
LLM Training Infrastructure · Agentic Systems · RAG & Retrieval · AI Observability
I'm an ML / AI research engineer working at the intersection of ML research and engineering. I came up through a PhD in computer science and peer-reviewed research, and I now spend my time on the parts that make models actually useful — distributed training and post-training, retrieval pipelines, multi-agent orchestration, and the observability to debug them. The projects below take research ideas and make them concrete; most come with code or a live demo so you can see exactly how they work.
I trained a small AI model that compresses speech down to roughly the size of a text message while keeping it understandable. Hit play on the left to hear the original recording, then play the right one — that's the same voice squeezed through the model and reconstructed. The same technology powers AI voice assistants like Moshi and VALL-E. Try it on your own voice — it runs entirely in your browser, your audio never leaves your machine.
About
I hold a PhD in computer science from UC Davis, where my research was in data and scientific visualization, with around ten peer-reviewed publications — including a Best Paper award. That research background shapes how I work: I'm comfortable with the literature and with rigorous evaluation, and I care about getting ideas from a paper or a prototype all the way to a system people can actually use.
These days that bridge runs through ML infrastructure. On the training side — DeepSpeed and FSDP pipelines, multi-GPU comparisons, and post-training sweeps with crash recovery. On the agent side — multi-agent orchestration and self-learning training loops with hierarchical judges. For retrieval — RAG with hybrid search, reranking, and citation traceability. And because these systems are hard to debug, I build observability tracers and judge panels — plus interactive visualizations that turn ML concepts into something you can explore directly.
Selected Work
A cross-section of agentic systems, training infrastructure, retrieval, and observability — research ideas built into working systems. Several are interactive: the GNN explorer, the live DeepSpeed training dashboard, the ML system design handbook and more all run in your browser. Look for the Live Demo button and open one — that's the fastest way to see how the work behaves. Open-source projects lead the grid; every card is marked public or private.
Research
Before ML infrastructure, I spent years doing peer-reviewed research — a PhD in Computer Science from UC Davis (with Lawrence Berkeley National Laboratory), focused on data and scientific visualization for neuroscience, networks, and deep learning. The work below is real, published, and includes a Best Paper award.
ACM-BCB · ACM Conference on Bioinformatics, Computational Biology & Health Informatics · 2016
BMC Bioinformatics · 2017
IEEE/ACM Transactions on Computational Biology and Bioinformatics · 2016
IEEE Computer Graphics and Applications
Cluster Computing · 2021
University of California, Davis & Lawrence Berkeley National Laboratory
Full publication list, teaching, and talks on the academic site.
Capabilities
Four areas, grounded in the projects above — the real technologies and patterns I reach for.
Get in touch
I work at the intersection of ML research and engineering — LLM training infrastructure, agentic systems, and retrieval. The fastest way to reach me is email.