I’m Junjie (JJ), a software engineer. I like taking ideas from rough prototype to something people can actually use, especially where product engineering and AI meet.
I’m currently building full-stack tools at Tencent. Previously, I shipped AI procurement products at Didero in New York and worked on multilingual model post-training at AI Singapore. More about that work is on my website.
Current rabbit hole: learning AI infrastructure, especially vLLM and how to serve models efficiently.
| Project | What it does |
|---|---|
| GrassDump | A Telegram-first capture system that sorts thoughts into notes and todos, with semantic search and reminders. |
| VoicesInMyHead | An in-progress, local-first macOS app for bilingual meeting notes and dictation. The code is private for now. |
| Bilingual sentiment prediction | Fine-tunes a small language model for English and Vietnamese sentiment classification. |
| Temperature and plant-stage prediction | Cleans agricultural sensor data and trains XGBoost models for regression and classification. |
| Project Eudaimonia | A Unity VR prototype for exploring animal phobias through interactive scenarios, built with a teammate. |
| Talkerinos | A Go and PostgreSQL backend for a small blog, with drafts, publishing, and typed SQL queries. |
| Grasshut | My little corner of the internet for projects, writing, and experiments. Built with Next.js and TypeScript. |
More experiments: QOLplus (legacy) · dotfiles
Usually tinkering with developer tools, AI workflows, and the small details that make software feel good to use.
