Epoch
Zerg / ZTC agent runtime
Autonomous coding-agent platform and terminal with parallel tool dispatch, background sub-agents, worktree-isolated batch execution, provider routing, and live web-mirror observability.
Michael Chen
builds
agents
that prove work.
AI software engineer at Epoch focused on autonomous coding-agent platforms, deterministic LLM evaluation, and long-running multi-agent simulations. Before that: fuel-cell test automation, firmware-adjacent systems, and UCR's first AI organization.
Current focus
Coding agents, eval harnesses, simulation runtimes
Engineering Surface
A portfolio built around systems that keep running after the demo.
Forward deployed AI
Continual validation
Embedded evaluation harnesses score outputs against deterministic state, tool, artifact, browser, and visual oracles.
MOBIVOLT
5,000+ hours
Automated fuel-cell test systems integrating 250+ sensors with C#/.NET, Python/PyQt, serial firmware bridges, and hardware-in-the-loop simulation.
AI@UCR
0 to 50+
Founded UCR's first official AI organization and built workshop curriculum, speaker programming, and onboarding infrastructure.
Selected Work
Public builds and private systems share the same bias: make the loop tighter.
ZTC
Zerg / ZTC
Agentic terminal infrastructure for coding agents, parallel tools, sub-agents, provider routing, and web-mirror observability.
Epoch AIPHI
PHiLIP
AMD award-winning human-in-the-loop image generation platform with fast inference, 71 style presets, and PixArt-alpha fine-tuning on ROCm.
Python / PyTorch82
Better 82-0
Next.js multiplayer Cap Mode with private lobbies, server-authoritative state, shared drafts, and production-like verification loops.
TypeScript / PostgresRD
ROCm-Doctor
Self-healing model CI/CD concept for AMD MI300X deployments, classifying drift versus operator changes and applying audited repairs.
PythonEV
Market models
Go and Python systems for NBA betting opportunities, dynasty projections, and decision support with model-backed ranking.
Go / PythonProof Loops
The through-line is instrumentation before confidence.
The resume reads like three versions of the same engineering instinct: build the system, build the harness around it, then keep tightening the feedback loop until failures become observable.
AI infrastructure
Evaluation harnesses with deterministic oracles
State, tool, artifact, browser, and visual checks turn AI behavior into something that can be regression-tested.
Hardware systems
Physical test loops with real sensors
Fuel-cell automation tied hundreds of sensors, serial protocols, plant models, and long-duration collection into one operating surface.
Community systems
Workshops, curriculum, and AI@UCR infrastructure
Technical programming and reusable materials helped students move from curiosity to applied AI contributions.
Operating Signal
Michael works where software has to touch messy reality.
Open Loop
Bring the hard problem with the messy constraints.
Best fit: coding-agent infrastructure, AI evaluation, simulation-heavy workflows, product systems that need real browser verification, or hardware-adjacent tooling where correctness has to be visible.