zhenlin.wang · /now

Live update by Claude + Codex

Machine Learning Engineer Agentic AI Systems Sunnyvale, CA

Writing & building agentic AI systems that reason, act, and hold up.

Machine learning engineer focused on production reasoning engines, agent memory, and event-driven RL. I turn ML research into deployable agentic systems: tool use, orchestration, evaluation, and robustness when conditions shift.

Focus Stack: Agent MemoryEvent-driven RLReasoning EngineML Robustness
01Writing
all writing ->
featured 2026·06·07

Agents Do Not Need To Train To Learn

The model weights can stay frozen while an agent learns through memory, rules, commands, skills, logs, and the workspace around it.

10 min
2026·06·07

The Missing Fence Between AI Plans And PR Stacks

AI agents can produce plausible plans quickly, but teams need a small review surface before those plans harden into a stack of PRs.

8 min
2026·05·25

Code Is Cheap, Show Me The Idea

AI made the first demo cheaper. The scarce thing now is the idea the model would not have found on its own.

7 min
02Selected publications
google scholar ->
NeurIPS 2023 · main

Characterizing Out-of-Distribution Error via Optimal Transport

Yuzhe Lu, Yilong Qin, Runtian Zhai, Andrew Shen, Ketong Chen, Zhenlin Wang, Soheil Kolouri, Simon Stepputtis, Joseph Campbell, Katia Sycara

papercode
34 cited
ICLR 2023 · workshop

Predicting Out-of-Distribution Error with Confidence Optimal Transport

Yuzhe Lu, Zhenlin Wang, Runtian Zhai, Soheil Kolouri, Joseph Campbell, Katia Sycara

papercode
26 cited
AISTATS 2022 · main

Best Arm Identification with Safety Constraints

Zhenlin Wang, Andrew Wagenmaker, Kevin Jamieson

paper
36 cited
03Projects
github ->

HERON

active

A domain-agnostic MARL framework for evaluating trained policies under heterogeneous event-driven execution and realistic observability constraints

multi-agent rlevent-drivensimulation
read

ResumeAssist

wip

A private, configurable AI resume assistant with LangChain agents, CV rendering, and a full-stack review workflow

agentic ailangchainfull-stack ai
read code ★ 1 · 80 commits · ◷ 1y

llm-validator

active

A configurable LLM benchmarking template for repeatable model, prompt, dataset, and metric validation

llm evaluationmodel validationbenchmarking
read codewriteup ★ 0 · 21 commits · ◷ 1y

dpai

archived

A lightweight local inference-serving toolkit for registering models and exposing prediction endpoints quickly

model servingmlopsfastapi
read codedocs ★ 1 · 16 commits · ◷ 2y