Howard
I’m Howard. I’m doing a bachelor’s in Information Engineering at TUM Campus Heilbronn in Germany, and I graduate in 2028.
I’m learning LLM inference, GPU programming, AI agents, and C++ and systems programming. This site is where I keep my projects and notes. It’s written in Chinese, except this page. When I post a number, I say which machine it came from and how to reproduce it.
What I’m looking for
An internship in AI, either infrastructure or agents.
- Starting mid-February 2027, for at least 3 months (exact dates depend on the TUM exam period)
- 5 days a week, on site
- Graduating in 2028
Projects
- deltabox (open-source sandbox for AI coding agents, Rust; ongoing). So far three merged PRs: concurrent multi-agent sessions with a publish lock, macOS session fixes, and the npm/Homebrew release pipeline. With 8 sessions racing on one file, rejected commits went from about 60 to 0 and p95 landing time from almost 2 s to about 185 ms. I work on it as part of a research internship at the IPADS lab, Shanghai Jiao Tong University, which started in August 2026.
- Pip, an LLM system for a desktop pet robot (8-person TUM course project). In the LLM group I worked on the first prototype, the LangGraph state machine, the web console and the voice pipeline.
- Automated QC platform for LLM training data (internship, April to August 2026). Rule checks plus LLM review through the official APIs; before reviewing a record it retrieves that record’s past revision notes and business-process reference. Results are routed by confidence, so people only review the uncertain ones, and their feedback is turned into rule suggestions for the next round.
- A study system for my semester, built with Claude: a four-layer knowledge base, 9 commands, a dashboard and an automatic morning brief.
- infinite-pi (C++, TUM course project). Pi to any number of digits with the Chudnovsky formula and binary splitting. At 30,000 digits, 8 threads are 6.40 times faster than one.
Each project page (in Chinese) says what my part was, where the numbers come from, and what didn’t work.
Contact
- Email: fine.pushing@gmail.com
- WeChat: PaperSiii
- GitHub: https://github.com/papersii
- Xiaohongshu: Papiersilikon (ID 95295201622)
If you get in touch, please tell me how you found this site.