# Hanif Leoputera

> Hanif Leoputera is a research engineer at Tacit in San Francisco, building wearables that enable silent speech. Previously at Affirm. This is his personal site and blog.

Every essay has a plain-markdown copy at the same URL with `.md` appended. Interactive figures are described by their captions there.

## Writing

- [Generative interfaces: a glimpse of the future](https://hanifleo.com/writing/generative-interfaces.md): AI is starting to draw interfaces. I've been wondering which part of the system should decide what you see.
- [Post Hot Chips 2026: a review of custom chips](https://hanifleo.com/writing/post-hot-chips-2026.md): A tour of custom AI chips, and why the interesting comparison is speed per user against throughput, not FLOPs.
- [Can LLMs do science?](https://hanifleo.com/writing/can-llms-do-science.md): A review of how Periodic, Lila, DeepMind, FutureHouse and academic labs are trying to make language models do science, and what still stands in the way.
- [A survey of the data market](https://hanifleo.com/writing/data-market.md): Why AI training data looks less like oil and more like fresh produce, and which parts of the market are worth building or backing.
- [A survey of speech-to-speech models](https://hanifleo.com/writing/speech-to-speech-survey.md): Who has built what in speech-to-speech, and how it works: four kinds of full duplex, codecs and continuous features, token factorizations, turn-taking decisions and thinking while listening.
- [Everything is evals](https://hanifleo.com/writing/everything-is-evals.md): A survey of formal proofs for AI-written code, from Lean, Dafny and Verus to Bend and the new AI provers, and why the proof has stopped being the hard part.

## About

- [Home](https://hanifleo.com/): bio, list of writing, contact
- Contact: hanifleoputeralim@gmail.com