Nicolas Perrin
CS student at CMU, previously EPFL. Mostly working on AI systems, databases, and whatever makes large models faster and less annoying to run.
About
I grew up in Lausanne, did my undergrad at EPFL, and I'm currently finishing a master's at CMU. I like systems that are simple on the outside and a bit clever on the inside.
Most of my time goes into the parts of machine learning that aren't the model itself — the storage, the query paths, the bottlenecks that only show up once things get big. I read a lot of source code and write the occasional small thing to see how it actually behaves.
Interests
- AI systems — serving, scheduling, and the unglamorous infrastructure underneath.
- Databases — query optimization, internals, and why things are slow.
- Efficient infrastructure for ML workloads — memory, throughput, latency.
Small projects
Mostly notes and experiments. Nothing finished, all in progress.
- petit-db A tiny toy database in Python — an append-only log, a small SQL parser, and a REPL. Built for reading, not for production.
- oh-la-llm Turns plain language into petit-db queries through an OpenAI-compatible API. A thin, honest translation step, nothing clever.
- Notes on query optimization Working through join ordering and cost models, one example at a time.
- Database internals reading notes Annotating storage engines and execution as I read through them.
Contact
Happy to chat about any of the above. Best reached by email or on GitHub.