Yufeng He · 何宇峰

Building reliableagent systemsfrom post-training to inference.

Former Moonshot AI (Kimi) researcher and current Mooncake OSS Maintainer, working across coding-agent post-training, agent harnesses, multi-agent systems, and LLM inference.

Agent harnessesCoding-agent post-trainingMulti-agent systemsLLM systems
View résumé ↗
Emailheyufeng2003@qq.com
GitHubgithub.com/he-yufengLinkedInlinkedin.com/in/yufenghe
M.Sc. Computer Science · The University of Hong Kong
Yufeng He
Agent systemsOpen sourceHong Kong
400+upstream PRs
60+inference-system fixes
200+agent-harness improvements
100+agent-application PRs

02 / Open source

Open source,
at systems depth.

400+ merged upstream PRs across leading Agent and inference projects, alongside independently developed Agent tooling.

Maintained & contributed upstream

Contributions across
the Agent systems stack.

400+merged upstream PRs
across 10+ leading AI projects
01

Agent harnesses

200+

Merged upstream PRs

Qwen Code ↗53+ upstream PRsCoding agent
OpenClaw ↗6+ upstream PRsAgent harness
DeerFlow ↗16+ upstream PRsAgent harness
02

Inference systems

60+

System-level fixes · Mooncake OSS Maintainer

vLLM ↗12+ upstream PRsServing & runtime
Mooncake ↗37+ upstream PRs · MaintainerKV-cache infrastructure
PyTorch ↗9+ upstream PRsCore systems
Triton ↗System contributionsGPU programming
03

Agent applications

100+

Merged upstream PRs

Vibe-Trading ↗31+ upstream PRs · MaintainerAgentic finance
AstrBot ↗30+ upstream PRsAgent application
Independent open source01

Creator & Maintainer

CoreCoder ↗

Independently developed; defined a core architectural pattern for Coding Agents.

1.7K+GitHub stars

Python · Runnable agent harness

01 / Experience

Research depth.
Engineering range.

Across agent harnesses, coding-agent post-training, multi-agent systems, agentic finance, and LLM infrastructure.

Kimi

Moonshot AI

AI Agent Researcher

Agent research that survives contact with engineering.

  • Owned exploratory experiments and engineering validation across agent post-training and harness frameworks at Kimi Alpha Lab.
  • Contributed to Kimi Coding Agent post-training alignment; independently built a post-training benchmark for long-horizon coding tasks and contributed to research on self-evolving agent post-training.
  • Researched and implemented multi-agent systems and agent-swarm methods; independently owned extensive multi-agent algorithm experiments.
  • Core contributor to a strategic quantitative-finance agents program; independently owned several key confidential agent projects.
01

Baidu

AI Agent Engineer
AI Talent Program

Developed an ad-bidding agent jointly modeling ROI and quality score, raising ROI target attainment from 81% to 101%.

02

Maimai

AI Agent Engineer
CEO Mentorship

Under CEO mentorship, independently built an agent pipeline for skill representation and quality evaluation, improving the automated career-recommendation loop.

03

Kuaishou

Machine Learning Engineer
Transcoding scheduling

Optimized transcoding allocation by jointly modeling content popularity, compute cost, and bandwidth value.

05 / Research

Studying how
agent systems fail.

My research focuses on evaluation blind spots: what long trajectories hide, how multi-agent systems break, and where benchmark scores stop telling the truth.

R/01

First author

TrajBias ↗

Built a coding-agent trajectory-evaluation framework and identified three classes of bias distinct from text evaluation.

Coding-agent trajectory evaluation

R/02

First author

FinAgentFail

Built a financial multi-agent failure benchmark with a nine-mode taxonomy.

Multi-agent systems · Financial agents

R/03

Sole author

FMGAN ↗

Applied adversarial refinement to time-series imputation, improving reconstruction stability.

Published · MiLeTS@KDD 2026

04 / Selected projects

Built beyond
the benchmark.

Two earlier systems that stretched from edge AI and digital twins to multi-agent music generation—and taught me how to own an engineering outcome.

CTO · Jun 2025 — Jul 202601

Edge AI · Health screening

FormulaSperm

Built an edge-AI health-screening system; trained video models and optimized on-device inference, achieving sub-3% digital-twin simulation error.

Full-Stack Engineer
Oct 2021 — Jun 202202

Generative AI · Music systems

Qiyin Tech

Pioneered a collaborative multi-agent music-generation engine that won National Gold at China's “Internet+” competition.

03 / Education

Sep 2025 — Nov 2026

The University of Hong Kong

M.Sc. in Computer Science

Sep 2021 — Jun 2025

Beijing University of Posts and Telecommunications

B.Eng. in Telecommunications Engineering

Selected honors

  • Champion2026 HKU DeepTech100 · HK$1M+ prize
  • Captain · 1st Place (Gold)BUPT · 2023 Shanghai Global AI Innovation Competition · LLM Track
  • 3× SilverACM International Collegiate Programming Contest (ICPC)
  • National First PrizeNational Olympiad in Informatics (NOIP)

03B / Skills & interests

Fluent across the stack.

Languages

Python / C++ (primary) · Java / Go

Agent systems

Agent harnesses · Coding-agent post-training · LLM systems

Practice

Claude Code / Codex power user · English (professional working proficiency)

Interest

National Top 10 · China Children's Go (Weiqi) Grand Prix