Idris Hanafi
1. Introduction
Hi, I’m Idris. I’m a product-minded systems engineer. I like building AI and LLM products that feel good to use, especially agentic ones, and I care a lot about making complex tech feel simple.
I’m happiest at the 0 to 1 stage: prototyping fast, figuring things out when nothing is clear yet, and then scaling it once it starts working. Along the way I’ve done full stack product engineering, scaled consumer products, and worked on distributed systems, trading systems, AI agents, and infra.
Most recently I was a founding engineer at The Grid, an exchange for buying and selling AI inference like a commodity, where I also built the agent harness for Syntax and launched an agentic hedge fund run by AI agents. Before that I worked on blockchain performance and developer tooling at Polygon Labs, co-founded Craftle to build an AI interior designer before the LLM hype, worked on large-scale data pipelines and infra at Amazon, and did full stack work at General Motors. On the side, I like to tinker with the latest tech, bike, hike trails, collect pokemon cards, and watch anime.
2. Side tinkering
2.1. wtm
A small Git worktree manager that copies your local env files into every new worktree, so you can run a team of coding agents on one repo, each on its own branch and opening its own PR.
Read the write-upMay 20262.2. Journey to fine-tuning a VLM to be good at computer use
Post-training a 500M vision language model on a MacBook to look at a screen and click the red button, from 0% to 92.5%, plus where a small computer use model goes from here.
Read the write-upSeptember 20262.3. Evaluate News: what other sites say about an article
Paste in a link, and it pulls out the main claims, finds similar articles, and uses RAG to show which sources agree or disagree, with quotes.
Read the write-upOctober 20242.4. …and many more that never got to see the light of day
Half-finished prototypes, weekend experiments, and ideas that lived and died in a scratch folder. Some of them might show up here eventually.
3. Currently
I’m exploring playful interfaces, AI-native tools, and fine-tuning local models for practical, everyday uses. This page is heavily WIP, so projects and notes will change as the work does.