---
document_type: resume
name: Leon Liu
alternate_name: 柳文浩
headline: Technical Developer Advocate - AI Agents - Evaluation - Open-Source Tooling
location: San Francisco Bay Area
languages:
  - English
  - Mandarin
email: leonwenhao@gmail.com
phone: 239-908-8278
website: https://www.leonwenhao.com/
linkedin: https://www.linkedin.com/in/leon-liu-555140142/
github: https://github.com/Leonwenhao
writing: https://leonliu.substack.com/
target_roles:
  - Technical Developer Advocate
  - Developer Advocate
  - Developer Relations Engineer
focus_areas:
  - AI agents
  - Agent evaluation
  - Open-source tooling
  - Developer education
  - U.S. and Chinese AI ecosystems
last_updated: 2026-07-18
canonical_url: https://www.leonwenhao.com/resume.md
pdf_url: https://www.leonwenhao.com/Leon_Liu_Technical_Developer_Advocate_Resume.pdf
---

# Leon Liu

**Technical Developer Advocate - AI Agents - Evaluation - Open-Source Tooling**

## Contact

- Location: San Francisco Bay Area
- Languages: English and Mandarin
- Email: [leonwenhao@gmail.com](mailto:leonwenhao@gmail.com)
- Phone: 239-908-8278
- Website: [leonwenhao.com](https://www.leonwenhao.com/)
- LinkedIn: [linkedin.com/in/leon-liu-555140142](https://www.linkedin.com/in/leon-liu-555140142/)
- GitHub: [github.com/Leonwenhao](https://github.com/Leonwenhao)
- Writing: [leonliu.substack.com](https://leonliu.substack.com/)
- PDF: [Leon Liu - Technical Developer Advocate Resume](https://www.leonwenhao.com/Leon_Liu_Technical_Developer_Advocate_Resume.pdf)

## Profile

Technical developer advocate and hands-on AI builder who turns real product use into open-source tools, tutorials, talks, and actionable product feedback. Achieved the highest OfficeQA accuracy in Sentient Arena Cohort 0 at $1.85 per run; builds agent-evaluation infrastructure and works natively across U.S. and Chinese AI ecosystems.

## Experience

### Founder & AI Builder - Dolores Research

- Dates: 2026 - Present
- Focus: Agent evaluation, open-source tooling, developer education
- Location: San Francisco and China

- Built Teller during [Sentient Arena Cohort 0](https://www.linkedin.com/posts/sentientagi_leon-liu-a-sentient-arena-builder-just-activity-7447846701673078785-nnON) and achieved the highest OfficeQA accuracy: 71.5% (176/246) at $1.85 per run with MiniMax M2.5. Sentient Arena is backed by Founders Fund, Pantera, and Franklin Templeton.
- Produced and presented an [EvoSkill setup tutorial](https://youtu.be/Yqutt4CuX5k) published by Sentient Labs, translating an agent self-improvement workflow into a reproducible developer walkthrough.
- Reported and isolated two Fireworks/OpenCode integration failures in [EvoSkill issue #48](https://github.com/sentient-agi/EvoSkill/issues/48), including a verified complete MiniMax run; Fireworks support shipped as a first-class provider in [v1.3.0](https://github.com/sentient-agi/EvoSkill/releases/tag/v1.3.0).
- Built Dolores Autocurricula, a local-first engine for generating, sandbox-verifying, solver-calibrating, scoring, and exporting software tasks; validated it through a 50-task, 250-attempt Fireworks calibration.
- Spoke at [Open AGI Builders Day](https://luma.com/4bi0owdl?locale=zh) (Alibaba x Sentient x Fudan) and was featured by Sentient Labs at [MU Shanghai](https://www.linkedin.com/posts/sentientagi_we-caught-up-with-arena-alum-leon-liu-at-activity-7464661555083898880-r0u1) discussing agent evaluation and China's open-source AI ecosystem.

## Selected Public Proof

### Teller

- Type: Open-source grounded reasoning agent
- Evidence: [GitHub repository](https://github.com/Leonwenhao/teller)
- Published the competition methodology, reusable agent configuration, and technical report for enterprise document QA.

### Yume

- Recognition: 1st place, Film & Simulation track - Worlds in Action
- Evidence: [Devpost](https://world-model-hackathon.devpost.com/)
- Built the backend and end-to-end generation pipeline for turning children's drawings into explorable 3D worlds in WebXR.

### Public Ecosystem Work

- Focus: Speaking, interviews, and media
- Evidence: [36Kr / 职场Bonus article](https://mp.weixin.qq.com/s/kW7WA7i-D6aOX_kEsuHaFw)
- Lead builder profiled in 36Kr's 职场Bonus (36Kr is comparable to China's TechCrunch) coverage of the muShanghai residency; represented open-source agent work across English- and Mandarin-speaking audiences.

## Prior Experience

### Product Lead & Business Development - Devermint

- Dates: December 2024 - December 2025
- Organization: Enterprise Web3 studio
- Location: Remote

- Led product strategy and go-to-market for Aptos AI Layer and SUI Agents, coordinating engineering, design, and ecosystem stakeholders.
- Authored technical documentation, grant applications, and developer guides used in the Aptos Foundation accelerator and ecosystem showcases.
- Built partnerships and ran developer- and user-facing communications, including AMAs, technical content, and launch messaging.

### Founder & Lead Product Builder - Anna Fantasy

- Dates: January 2025 - November 2025
- Organization: Generative AI video platform
- Location: San Francisco

- Designed and shipped an AI video platform with structured prompting, camera-scene controls, and community feedback loops; grew to 3,000+ monthly active users across 50+ countries.
- Turned complex generative-video workflows into product language and educational content for non-technical creators; generated 500K+ organic X impressions.

### Head of Communications - DeVol Network

- Dates: April 2022 - May 2023
- Organization: Decentralized options protocol on Solana
- Location: New York

- Supported a $3.5M seed raise through investor materials and outreach; built partnerships with Polygon, Circle, and Solana and translated DeFi concepts into technical and community content.

### Web3 Innovation Consultant - Nike, Inc.

- Dates: June 2021 - October 2021
- Location: Remote

- Advised senior Nike stakeholders on NFT strategy, blockchain-based digital collectibles, and the CryptoKicks initiative.

## Additional Open-Source Project

### DeepRepo

- Type: Recursive codebase analysis
- Evidence: [GitHub repository](https://github.com/Leonwenhao/deeprepo)
- Built a CLI that coordinates a root model and focused worker models to analyze repositories and produce reusable technical context for coding agents.

## Selected Writing

- [The Idea Guy Learned to Ship](https://leonliu.substack.com/p/the-idea-guy-learned-to-ship)
- [Post-Training Is Just Dog Training](https://leonliu.substack.com/p/post-training-is-just-dog-training)
- [Chinese AI with American Characteristics](https://leonliu.substack.com/p/chinese-ai-with-american-characteristics)

## Capabilities

### Developer Relations

- Technical tutorials
- Demos
- Documentation
- Talks and workshops
- Product-feedback loops
- Community and ecosystem partnerships

### AI & Evaluation

- Agent harnesses
- Prompt and behavioral engineering
- Benchmark workflows
- Solver panels
- Verifier design
- Open-model inference

### Tools

- Python
- JavaScript and TypeScript
- React and Next.js
- Git
- Docker
- CLI tooling
- CI/CD
- Fireworks
- OpenRouter
- Claude Code
- Codex

## Education

### Emerson College

- Field: Visual & Media Arts / Film Production