CV
Weijia (Charlie) Zhang
Incoming M.S. in Computer Science @ Yale
Summary
Incoming Yale M.S. student in Computer Science, UIUC Computer Science and Mathematics graduate, and NLP/LLM agent researcher.
Education
- M.S. in Computer Science2028-05Yale University
- B.S. in Computer Science and Mathematics2026-05University of Illinois Urbana-ChampaignGPA: 3.7/4.0
Work Experience
- Machine Learning Engineer Intern, RSI / Self-Evolving AgentsJun 2026 - PresentTikTokProject lead for TikTok's RSI/Self-Evolving Agent framework, built from 0 to 1.
- Designed and shipped a closed Solver-Reflector-Evolver loop in which the Solver executes tasks, the Reflector attributes failures, and the Evolver iterates reusable skills; lifted F1 from 62.7% to 81.3% on unoriginal-video detection (9,600 real short videos) and from 76.5% to 87.3% on reposting-account detection (200 accounts).
- Designed a transferable Online Failure Discovery process: the Reflector mines failure patterns from scenarios and Solver traces, maintains a two-level scenario-to-failure-mode taxonomy, and distills high-value error clusters into reusable skills, so the same framework self-evolves across scenarios without human-predefined error types.
- Trained a video-understanding model: built an account-level SFT dataset of 74K accounts and 1.33M videos, ran distributed full-parameter SFT of Qwen3-VL 2B/4B/8B on 32 H100s with DeepSpeed, and calibrated thresholds across 53 checkpoints; the final 4B model cut false positives by 42.8% (318 to 182) against the 8B baseline at ~70% recall.
- Research Intern, Large Language ModelsJul 2025 - Sep 2025MicrosoftBuilt the Excel Coding Agent data engine and post-training pipeline for Excel Copilot's code agent.
- Produced execution-verified SFT/RL data by mining 20,000+ real-world workbooks, back-translating user queries from spreadsheet context, sampling multi-turn Office.js rollouts (write, execute, repair), and using SheetEngine final-state assertions as a programmatic verifier to filter incorrect and reward-hacking solutions.
- Post-trained the coding agent on this corpus with rejection-sampling SFT and RL over same-task pass/fail rollouts under execution-based reward, improving Office Scripts pass@1 by 15% against the production baseline on a held-out set.
- Software Development InternAug 2024 - Sep 2024Tencent, WeChat Group (WXG)Built cross-platform performance analysis and Unity memory-profiling tools for WeChat Mini Programs.
- Developed a cross-platform Android and iOS hardware performance analysis tool with WeChat Mini Program telemetry, helping 200+ partner teams identify performance bottlenecks.
- Built a Unity Mono Memory Profiler that identified 40+ hidden memory allocation sites in a pilot application.
- AI Agent EngineerMay 2023 - Jul 2023Reborn NetworkBuilt real-time embodied role-playing agents in Unity VR games.
- Built an embodied role-playing agent in a Unity VR environment, closing a real-time perception-dialogue-action loop across text, voice, and full-body VR actions at sub-second end-to-end latency.
- Introduced RAG/vector databases and dual-level (episodic + semantic) memory, improving cross-session recall accuracy from 38% to 61% on an internal multi-session dialogue eval.
Skills
Programming Languages
- Python
- C/C++
- C#
Frameworks and Libraries
- VERL
- vLLM
- LangGraph
- LangChain
- PyTorch
- TensorFlow
AI Focus
- RL/Post-training
- LLM Agents
- Multimodal Reasoning
- Self-Evolving Agents
- SFT
- RAG
- Evaluation
Publications
- How Much Vision Does Multimodal Reasoning Need? Vision-Stripping for Multimodal Benchmarks2026Under review at NeurIPS 2026Research on vision-stripping for multimodal reasoning benchmarks.
- Every Act Has Its Price: Compressed Moral Composition in Frontier LLMs2026arXiv preprint arXiv:2606.11232Research on compressed moral composition in frontier LLMs.
- SeeingEye: Agentic Information Flow Unlocks Multimodal Reasoning in Text-only LLMs2025arXiv preprint arXiv:2510.25092; under review at AAAI 2026Agentic information flow for multimodal reasoning in text-only LLMs.
- CUADebug: Diagnosing and Repairing Computer-Use Agent Failures2026
- Artificial Leviathan: Exploring Social Evolution of LLM Agents Through the Lens of Hobbesian Social Contract Theory2024The First Workshop on AI Behavioral Science, ACM SIGKDD 2024A simulated LLM agent society for studying emergent social contracts.
- Understanding and Mitigating the Bias Inheritance in LLM-based Data Augmentation on Downstream Tasks2026
- Where LLM Agents Fail and How They Can Learn From Failures2025arXiv preprint arXiv:2509.25370; Failure Modes in Agentic AI Workshop, ICML 2026Research on LLM agent failures and learning from failed trajectories.
Portfolio
- OpenManus & OpenManus-RL2025ResearchCore author of OpenManus-RL, a verl-based agent RL post-training framework in an open-source ecosystem with 70,000+ GitHub stars.
- SeeingEye2025ResearchAgentic information-flow framework that enables text-only LLMs to perform multimodal reasoning through structured, tool-mediated textual context.
- AgentDebugX2026ResearchComprehensive debugging framework for long-horizon agent trajectories, with a Detect-Attribute-Recover-Rerun loop across multi-agent, tool-using, and computer-use agents.
Interests
- ResearchLLM agents, VLM/LLM agents, Multimodal reasoning, Agent debugging, Reinforcement learning