I design and ship Agentic AI systems at web-scale...
Ph.D. in Computer Science from Télécom SudParis / Université Pierre et Marie Curie, France (2015). Postdoc at University of Virginia, then tenure-track Assistant Professor at Missouri S&T. Since 2018, held technical staff roles at Baidu and Microsoft. This is a personal site — opinions are my own.
Recurring design decisions around AI agents — what may vary, what must hold, and what happens when it does not. Published as an open field guide and a working agent package.
A field guide to 52 harness design records: UML class views, object interactions, pointcuts, protected contracts, permitted variation, evidence, and lifecycle questions.
xhyccc.github.io/pohe ↗ Open agent packageAn agent, CLI, and three skills that detect pattern cues in an arbitrary repository, pull structural candidates into source review, and plan pattern-guided harness evolution without losing the protected contract.
xhyccc.github.io/pohe/agent.html ↗Autonomous AI agents with LLM-powered function calling, tool use orchestration, and multi-step reasoning.
From learning theory and statistical data sciences to learning-to-rank systems at web-scale.
Production ML platforms serving billions of requests — model-as-a-service, distributed GNN training, multi-GPU graph learning.
Deep learning for scientific problems — RNA language models, molecular pre-training, protein-ligand binding, PDE surrogates.
Always interested in hard problems — agentic AI, search, and systems that scale.