Agentic AI Researcher @ Moonshot AI (Kimi) | MS CS @ HKU | Champion, Shanghai Global AI Contest | 3x ACM-ICPC Silver Medalist | Former Intern @ Baidu, Maimai, Kuaishou
| Project | Stars | What it does |
|---|---|---|
| MTSIR3-GAN | 135+ | Adapting R3GAN (NeurIPS 2024) to multivariate time series imputation + adversarial refinement study |
| FindJobs-Agent | 97+ | LLM-powered job toolkit: skill gap analysis, mock interviews, resume optimization |
| DRL-MultiFactorTrading | 67+ | Deep RL trading strategies with Double DQN and multi-factor alpha signals |
| Arxiv Paper Coding Agent | 36+ | Multi-LLM crew that reads a paper and produces a working implementation |
| ContractGuard | 43+ | AI agent that reviews contracts for red flags before you sign |
| AgentProbe | 33+ | Pytest plugin for regression-testing AI agents: snapshot baselines, semantic comparison, mock LLMs |
| PromptDiff | 25+ | Semantic diff for LLM prompts, compare prompt versions like git diff |
| TokenTracker | 14+ | Drop-in LLM cost tracker, change one import line, see where your money goes |
| GitSense | 13+ | AI-powered open source contribution finder, tell it your skills, it finds issues |
| RuleForge | 11+ | Auto-generate AI assistant rules (CLAUDE.md, .cursorrules) from codebase analysis |
| BatchLLM | 8+ | Batch LLM processing: CSV in, processed CSV out, with retries and cost tracking |
| Project | PR | What I Fixed |
|---|---|---|
| Mooncake (5k★) | #1629 | GB200 MNNVL EP hang: cudaMalloc → cuMemCreate(FABRIC) + cuMemMap for cross-node NVLink |
| Mooncake (5k★) | #1728 | Hard pin for eviction-protected objects: model weights never get evicted, const field + BatchEvict skip + backward-compat serialization |
| Mooncake (5k★) | #1644 | MNNVL warmup hang: skip redundant handshake for fabric-connected nodes |
| Mooncake (5k★) | #1626 | Silenced error log spam for non-memory replicas in metadata store |
| vLLM (74k★) | #37884 | RoBERTa position_ids in-place accumulation on CUDA graph padding: BGE-M3 crash after ~4000 requests |
| vLLM (74k★) | #37301 | Base64 JPEG video frames returning empty metadata: populate frame count, fps, duration |
| Transformers (158k★) | #44710 | Fix AutoProcessor.from_pretrained silently dropping hub kwargs (revision, token, etc.) |
| FlashInfer (5k★) | #2756 | Fix autotuner crash when input tensor is None: proper None-checking for optional inputs (fixes #2749) |
| AstrBot (28k★) | #6596 | Multimodal token counting: images, audio, chain-of-thought were invisible to context compression |
| AstrBot (28k★) | #6474 | SQLite database is locked under concurrent writes: added busy timeout |
| AstrBot (28k★) | #7003 | SSE heartbeat for WebChat: long context compression killed the connection |
| AstrBot (28k★) | #6581 | Context truncation dropping the only user message: causes 400 from Zhipu/Gemini |
| AstrBot (28k★) | #6656 | /stop follow-up race: agent_stop flag not checked during follow-up capture |
| AstrBot (28k★) | #6710 | Skills-like re-query dropping image captions: extra_user_content_parts not forwarded |
| AstrBot (28k★) | #6527 | Fix LLM tool selection: rewrote ambiguous Upload/Download descriptions |
| AstrBot (28k★) | #6313 | Null choices guard: OpenAI API returning None instead of empty list |
| AstrBot (28k★) | #7217 | Fix qwen3-rerank response parsing: handle both old and new Dashscope API formats |
| ClawTeam (4.1k★) | #1 | First PR: 122 tests, CI, team templates, config bugfixes, task duration tracking |
| ClawTeam (4.1k★) | #40 | Pluggable TaskStore: extract task persistence into swappable backend abstraction |
| ClawTeam (4.1k★) | #32 | Gemini CLI support: spawn, permissions, prompt injection for both backends |
| ClawTeam (4.1k★) | #36 | Kimi CLI support: spawn backend, permission handling, 3 new test cases |
| ClawTeam (4.1k★) | #24 | Team snapshot/restore: checkpoint and recovery for long-running agent swarms |
| LightRAG (31k★) | #2796 | Fix None file_path propagating as unknown_source: fill gaps left by #2793 |
| OpenClaw (343k★) | #41271 | Log auth profile resolution failures instead of swallowing silently |
LinkedIn: https://www.linkedin.com/in/yufenghe
Moonshot AI (Kimi) AI Agent 研究员 | 港大计算机硕士 | 上海全球AI大赛冠军 | 三次获ACM-ICPC银牌 | 曾在百度、脉脉、快手的AI 研发岗实习
| 项目 | Stars | 简介 |
|---|---|---|
| MTSIR3-GAN | 135+ | 将 R3GAN (NeurIPS 2024) 适配到多变量时序插补 + 对抗精炼实证研究 |
| FindJobs-Agent | 97+ | LLM 求职工具箱：技能差距分析、模拟面试、简历优化 |
| DRL-MultiFactorTrading | 67+ | 深度强化学习量化交易，Double DQN + 多因子 Alpha |
| Arxiv Paper Coding Agent | 36+ | 多 LLM 协作 Agent，读论文自动生成可运行代码 |
| ContractGuard | 43+ | AI 合同审查 Agent，签字前帮你找出霸王条款 |
| AgentProbe | 33+ | AI Agent 回归测试 pytest 插件：快照基线、语义比较、Mock LLM |
| PromptDiff | 25+ | LLM prompt 语义 diff 工具，像 git diff 一样对比 prompt 版本 |
| TokenTracker | 14+ | LLM 开销追踪，改一行 import 就能用 |
| GitSense | 13+ | AI 驱动的开源贡献发现器，输入技能自动找 issue |
| RuleForge | 11+ | 从代码库自动生成 AI 助手规则（CLAUDE.md、.cursorrules） |
| BatchLLM | 8+ | LLM 批量处理：CSV 进 CSV 出，自动重试 + 费用追踪 |
| 项目 | PR | 修了啥 |
|---|---|---|
| Mooncake (5k★) | #1629 | GB200 MNNVL EP hang：cudaMalloc → cuMemCreate(FABRIC) + cuMemMap 跨节点 NVLink 通信 |
| Mooncake (5k★) | #1728 | Hard pin 驱逐保护：模型权重永不被驱逐，const 字段 + BatchEvict 跳过 + 向后兼容序列化 |
| Mooncake (5k★) | #1644 | MNNVL warmup hang：跳过 fabric 连接节点的冗余握手 |
| Mooncake (5k★) | #1626 | 修复非内存副本的错误日志刷屏 |
| vLLM (74k★) | #37884 | 修复 RoBERTa position_ids 原地累积溢出：BGE-M3 约 4000 请求后 CUDA graph padding crash |
| vLLM (74k★) | #37301 | 修复 base64 JPEG 视频帧返回空 metadata：补充帧数、fps、时长 |
| Transformers (158k★) | #44710 | 修复 AutoProcessor.from_pretrained 静默丢弃 hub kwargs（revision、token 等） |
| FlashInfer (5k★) | #2756 | 修复 autotuner 在输入 tensor 为 None 时崩溃（fixes #2749） |
| AstrBot (28k★) | #6596 | 多模态 token 计数：图片/音频/思考链对 context 压缩不可见的问题 |
| AstrBot (28k★) | #6474 | 修复 SQLite 并发写入 database is locked，添加 busy timeout |
| AstrBot (28k★) | #7003 | 修复 WebChat 长响应断连：SSE 心跳保活，context 压缩期间不再超时 |
| AstrBot (28k★) | #6581 | 修复截断器丢失唯一 user 消息导致智谱/Gemini 返回 400 |
| AstrBot (28k★) | #6656 | 修复 /stop 后新消息仍被 follow-up 捕获的竞态条件 |
| AstrBot (28k★) | #6710 | 修复 skills-like re-query 丢失图片描述：extra_user_content_parts 未传递 |
| AstrBot (28k★) | #6527 | 修复 LLM 工具选择：重写模糊的 Upload/Download 描述 |
| AstrBot (28k★) | #6313 | 修复 OpenAI API 返回 None choices 导致崩溃 |
| AstrBot (28k★) | #7217 | 修复 qwen3-rerank 响应解析：兼容新旧百炼 API 格式 |
| ClawTeam (4.1k★) | #1 | 首个 PR：122 个测试、CI、团队模板、config 修复、任务耗时追踪 |
| ClawTeam (4.1k★) | #40 | 可插拔 TaskStore：将任务持久化抽取为可替换的后端抽象层 |
| ClawTeam (4.1k★) | #32 | Gemini CLI 支持：spawn、权限、prompt 注入双 backend 适配 |
| ClawTeam (4.1k★) | #36 | Kimi CLI 支持：spawn backend、权限处理、3 个新测试 |
| ClawTeam (4.1k★) | #24 | 团队快照/恢复：长时间 agent swarm 的检查点和恢复机制 |
| LightRAG (31k★) | #2796 | 修复 None file_path 传播为 unknown_source：补 #2793 遗漏的处理层 |
| OpenClaw (343k★) | #41271 | 认证配置解析失败时记录日志而非静默吞掉 |