Weco AI
Weco is a research and product lab building recursively self-improving AI. Its product automates the scientific method for software: agents propose code changes, run evaluations, retain improvements, and iteratively optimize against measurable goals.
At a Glance
- Machine-learning and data-science teams
- AI startups and engineering teams
- Frontier AI research labs
- Financial services and fraud detection teams
- +2 more
AI Tools by Weco AI
(1)
Weco
Metric Driven Code Optimization CLI
Discussions
No discussions yet
Be the first to start a discussion about Weco AI
Latest News
Products & Services
A production-grade autoresearch engine that takes a codebase and evaluation script, uses LLM-guided tree search to propose and test variants, and keeps the best-performing version against a user-defined metric.
Open-source command-line optimizer installable with pip or pipx. It runs evaluation-driven code optimization, supports multi-file sources, and includes setup skills for Claude Code and Cursor plus an Observe mode for tracking external experiments.
Open-source LLM agent for machine-learning engineering and the research foundation from which Weco grew. It autonomously designs, implements, evaluates, and refines data-science solutions using solution-space tree search.
Dashboard observability for autoresearch runs, including experiment trees, code diffs, terminal output, metric tracking, and logging for externally controlled agents or workflows.
Market Position
Weco positions itself as production-grade, traceable, steerable autoresearch for arbitrary code and any measurable objective, rather than a conventional AutoML tuner or one-shot code generator. Its public materials distinguish it from Optuna and Claude Code's autoresearch approaches and frame it as a practical, broadly available alternative to research systems such as Google DeepMind's AlphaEvolve, with an open-source AIDE lineage and human-in-the-loop controls.
Leadership
Founders
Zhengyao Jiang
Computer scientist; PhD in machine learning from University College London. Co-authored the AIDE research and is Weco's co-founder and CEO.
Yuxiang Wu
Computer scientist; completed a PhD in Computer Science at University College London and previously earned a research-based MPhil in Computer Science and Engineering from the Hong Kong University of Science and Technology. Co-founder and CTO of Weco.
Executive Team
Zhengyao Jiang
Co-founder & CEO
Computer scientist with a PhD in machine learning from UCL; co-author of AIDE and focused on LLM-driven agents for automated ML research and code optimization.
Yuxiang Wu
Co-founder & CTO
Computer scientist with a UCL Computer Science PhD and HKUST research master's degree; leads Weco's technical architecture and AIDE.
Founding Story
Weco grew out of the founders' AIDE research, which used LLM-driven agents and solution-space tree search to automate machine-learning engineering. They started Weco to turn that research into evaluation-driven coding and make expert-grade, evidence-backed software optimization available to engineers and organizations, rather than limiting self-improvement to large research labs.
Business Model
Revenue Model
Usage-based hosted autoresearch credits, bring-your-own-key usage, and tailored enterprise optimization, deployment, and support. The open-source CLI is also distributed through pip/pipx.
Pricing Tiers
20 free credits, approximately 100 steps of Weco-hosted autoresearch; includes Weco Observe access.
Use OpenAI, Anthropic, or Gemini keys; up to 10,000 experiments per month.
Optimization and evaluation for teams, with deployment and support tailored to the workflow.
Target Markets
- Machine-learning and data-science teams
- AI startups and engineering teams
- Frontier AI research labs
- Financial services and fraud detection teams
- Researchers and scientific organizations
- Developers optimizing prompts, agents, models, and production code
- Agent and harness engineering
- Prompt engineering
- Machine-learning model and pipeline development
- Hyperparameter optimization
- GPU kernel and inference optimization in CUDA and Triton
- Fraud detection and financial models
- Vardera Labs
- OpenAI
- Standard Chartered
- Arbium