I build tools to make AI agents useful, inspectable, and measurable — from retrieval and research workflows to reproducible benchmarks and interactive evaluation. I also contribute correctness fixes and regression tests to open-source LLM inference engines.
B.Sc. student in Information and Computing Science at Beijing University of Posts and Telecommunications (BUPT), graduating in 2027. Admitted to the master’s program at BUPT’s School of Computer Science, starting in 2027. Open to remote AI / agent engineering and research internships.
LinkedIn · Deep Research Arena · Play LLM Arcade
10 merged PRs · 6 upstream repositories
| Repository | Merged PRs | Recent PRs |
|---|---|---|
| livekit/agents | 1 | #8571 |
| vllm-project/vllm | 4 | #68055 · #67904 · #67813 |
| ag2ai/ag2 | 1 | #3684 |
| huggingface/transformers | 2 | #50758 · #50763 |
| deepset-ai/haystack-core-integrations | 1 | #4268 |
| sgl-project/sglang | 1 | #44861 |
Updated every 30 minutes.
All merged public PRs · Open public PRs
| Project | What to explore |
|---|---|
| Deep Research Arena | A team project for evaluating deep-research agents in controlled web sandboxes. Citation reachability and quote matching complement pairwise quality judgments and Bradley–Terry / Elo ranking. Live arena → |
| ScholarMind | A research-workflow application with nine modules, live progress updates, traceable execution, pause/resume controls, and human review. Built with FastAPI, SQLAlchemy, WebSocket, React, and TypeScript. |
| LLM Arcade | Playable artifacts for evaluating models and coding harnesses. Visitors can try the results, submit full rankings, and vote in anonymous A/B comparisons. Try the arcade → |
| ReadTimeMemory | Core implementations for stateful agent memory, evidence retrieval and reranking, with recorded development results. Previously named stateful-memory; one project, one repository. |
More research code: HybridAttn · CEDAR.
Python · PyTorch · FastAPI · asyncio · SQLAlchemy · React · TypeScript · Docker · Git
My interests include agent evaluation, retrieval and grounding, research-workflow orchestration, and reliable tool/API integration. I care about reproducible results, clear failure modes, and interfaces that people can actually use.
Looking for a collaborator or intern? Reach out on LinkedIn. I’m particularly interested in teams building AI agents, evaluation infrastructure, and research tools.
