I build autonomous research agents at Weco AI, where I'm a Member of Technical Staff. In 2026, an agent I built produced seven accepted leaderboard records in OpenAI's Parameter Golf. OpenAI featured one of the agent's model-compression submissions in its competition retrospective, crediting my GitHub account, dexhunter. The agent writes code, tests it against a metric, and keeps improving it through long research runs without supervision.
I co-authored the AIDE paper (2025) and have contributed to the agent's codebase since 2024. AIDE uses tree search to write and improve machine learning code. OpenAI used AIDE for its MLE-bench evaluations of GPT-4.5, o1, and o3-mini, as documented in the GPT-4.5 system card. Meta FAIR's AI Research Agents described AIDE as "the state-of-the-art approach" and rebuilt it as a baseline for comparison.
My contributions to Inspect, the UK AI Security Institute's open-source LLM evaluation framework, include merged improvements that reduce scoring time and memory use and make tool-result media extraction scale linearly with conversation length. I was also a maintainer of the Hyperledger Fabric Python SDK, a Linux Foundation project.
Previously, I built trading backends in Rust, Go, and Node.js at Hex Trust. I studied Information and Computing Sciences at the University of Liverpool and Xi'an Jiaotong-Liverpool University, with earlier research at Nanyang Technological University, Zhejiang University, and Hong Kong Baptist University.
Featured
- UK AI Security Institute — Inspect — performance work on the UK government's LLM evaluation framework: cut clustered-stderr scoring time and memory (#4714), and made tool-result media extraction linear in conversation length (#4628).
- OpenAI Parameter Golf — I built an autonomous research agent that set seven leaderboard records, with entries in the official record-track directory. Its best result achieved a 5-seed mean validation BPB of 1.0645. The agent's submissions were made through my GitHub account and appear in the table below.
- Agent runtimes — merged performance work into openclaw, AutoGPT, goose, qwen-code, pydantic-ai, agno and BAML.
- AI research infrastructure — cut redundant AST parsing in Sakana AI's ShinkaEvolve and bounded process-pool shutdown latency in OpenEvolve; smaller docs fixes in Meta's aira-dojo, Microsoft's RD-Agent and OpenAI's MLE-bench.
- AIDE — contributions to the open-source tree-search agent behind my 2025 paper; it writes, evaluates, and improves machine learning code. I also contribute to weco-cli, the command line tool that drives it.
Every project below links to its merged pull requests on GitHub, so anything here can be checked directly. Ranked by stars, refreshed weekly.
Contributions to 52 open source projects — 29 of them AI or agent infrastructure.
| Project | Stars | Contributions | Latest |
|---|---|---|---|
 openclaw/openclaw |
389k | View PRs | Jul 2026 |
 Significant-Gravitas/AutoGPT |
187k | View PRs | Jul 2026 |
 langchain-ai/langchain |
146k | View PRs | Mar 2024 |
 OpenBB-finance/OpenBB |
73k | View PRs | Jul 2026 |
 aaif-goose/goose |
54k | View PRs | Jul 2026 |
 agno-agi/agno |
42k | View PRs | Jul 2026 |
 invoke-ai/InvokeAI |
28k | View PRs | Aug 2026 |
 QwenLM/qwen-code |
28k | View PRs | Aug 2026 |
 pydantic/pydantic-ai |
20k | View PRs | Jul 2026 |
 lllyasviel/style2paints |
18k | View PRs | Aug 2017 |
 pipecat-ai/pipecat |
15k | View PRs | Jul 2026 |
 microsoft/RD-Agent |
14k | View PRs | Sep 2025 |
 tadata-org/fastapi_mcp |
12k | View PRs | Mar 2025 |
 phillipi/pix2pix |
11k | View PRs | Jun 2017 |
 BoundaryML/baml |
9.1k | View PRs | Aug 2026 |
 algorithmicsuperintelligence/openevolve |
7.3k | View PRs | Jul 2026 |
 openai/parameter-golf |
5.2k | View PRs | Apr 2026 |
 TanStack/ai |
3.1k | View PRs | Aug 2026 |
 UKGovernmentBEIS/inspect_ai |
2.7k | View PRs | Aug 2026 |
 ZhengyaoJiang/PGPortfolio |
1.9k | View PRs | Dec 2017 |
9 more AI projects
| Project | Stars | Contributions | Latest |
|---|---|---|---|
| openai/mle-bench |
1.7k | View PRs | Nov 2025 |
 Farama-Foundation/ChatArena |
1.6k | View PRs | Jun 2023 |
 WecoAI/aideml |
1.5k | View PRs | Jul 2026 |
 SakanaAI/ShinkaEvolve |
1.4k | View PRs | Aug 2026 |
 tongjingqi/AI-Can-Learn-Scientific-Taste |
432 | View PRs | Mar 2026 |
 facebookresearch/aira-dojo |
165 | View PRs | Jul 2025 |
| WecoAI/weco-cli |
94 | View PRs | Sep 2025 |
 JeanKaddour/sokoban_speedrun |
31 | View PRs | Jul 2026 |
 openbydesign/lush |
3 | View PRs | Aug 2026 |
23 projects outside AI
- AIDE: AI-Driven Exploration in the Space of Code (arXiv), arXiv preprint, 2025
- Lightweight and Unobtrusive Data Obfuscation at IoT Edge for Remote Inference (DOI), IEEE Internet of Things Journal, 2020
- Challenges of Privacy-Preserving Machine Learning in IoT (DOI), ACM AIChallengeIoT, 2019
- A Deep Reinforcement Learning Framework for the Financial Portfolio Management Problem (arXiv), arXiv preprint, 2017
Citation counts are on Google Scholar.
- Hands-on AutoResearch: Cracking OpenAI's Parameter Golf — workshop with the Weco AI team, AI Engineer World's Fair 2026
- Algorithmic Trading Workshop — Network School, first cohort (2024)
- Deep Learning for Power System Security Assessment (2019)
- Introduction to Hyperledger Fabric (2019)
- 🏆 Special Prize (US$10,000), Wanxiang Blockchain Hackathon by QTUM (2018)
- 🥇 1st Prize, EOS Hackathon Hangzhou (team, 2018)
- 🥇 1st Prize, Hack x FDU 2017 Hackathon (out of more than 70 teams)
- 🥈 2nd Prize, XJTLU Blockchain Technology Application Innovation & Entrepreneurship Challenge (2020)
- 🥈 2nd Prize, XJTLU & PNP AI Innovation Hackathon (2018)
- 🥉 3rd Prize, EOS Hackathon Hangzhou (individual, 2018)
- 🥉 3rd Prize, DoraHacks x BCH Faith Hack (2018)
- 🏆 IBM Student Innovation Lab Program Award (2017)
- 🎓 Hyperledger Diversity Scholarship, Hyperledger Global Forum (2020)
- 🎓 CNCF Diversity Scholarship, KubeCon + CloudNativeCon China (2018)
An open-source tool I built: WakaTime-style usage tracking for Claude Code, Codex, and OpenCode. The charts below are my own usage, refreshed daily.
