About
I’m Wei. I’m a Staff ML engineer at Meta, where I work on multimodal LLMs and large-scale recommendation systems.
Most of my career has been about shipping ML at billion-user scale — ranking, retrieval, multimodal models, and the infra that holds it all together. I started in academic ML (PhD in CS at UW–Madison with David Page, focused on causal inference and healthcare ML), then moved to industry to build systems that touch a lot of people.
Lately I’ve been spending more time thinking about the frontier: alignment-adjacent questions, post-training and RLHF (which has a lot more in common with rec-sys than people usually admit), and the architectural choices that will shape the next few generations of foundation models.
What you’ll find here
- Deep dives — long-form posts drawn from production experience: scaling, ranking, multimodal, RLHF, the messy parts.
- Research notes — reactions to papers and releases I’ve been reading. Closer to a notebook than a review.
- Short takes — links + a paragraph or two. The Simon-Willison-style format that keeps the site alive between long pieces.
No paid tier, no comments, no popups. The RSS feed is the canonical way to follow along.
Elsewhere
- Code: github.com/razhangwei
- Publications: Google Scholar
- Professional: LinkedIn
- Email: razhangwei@gmail.com
Colophon
Built with Astro using the AstroPaper theme. Source on GitHub. Hosted on GitHub Pages.
Opinions are mine, not my employer’s.