Why the future of AI automation isn't computer-use agents, but RPA workflows compiled from screen recordings.
Hey, I'm Vidyoot
I'm currently building Granite, a platform that compiles screen recordings into deterministic, self-healing automations. I like building computer-use agents, shipping cool features, writing Rust code, doomscrolling Twitter, and occasionally writing blog posts.
I'm also a CS student at Georgia Tech. I've spent the last few years at early-stage startups building LLM agents, search infra, and dev tools.
Work Experience
- Ergo (YC W25) — Built LLM agents for sales automation. Website
- Mintlify (YC W21) — Built a Slack support agent used across 120+ workspaces. Website
- Trieve AI (YC W24) — Built search infrastructure powering 25k+ documentation sites. Website
- Caterpillar Inc. — Built 3D vision analytics to measure box space and optimize packing.
- University of Chicago — Engineered cloud-based protein simulation tools with 6x speedup.
Projects
- Granite — Computer-use agent that compiles UI workflows into deterministic automations. Website
- Threshold Securities — Multi-sig threshold Bitcoin wallet using Rust and libp2p. Repo
- JHMC Testing Portal — Full-stack platform serving 500+ math competitors annually. Repo
- Dynalab — Cloud-based protein simulation using Python and C++. Repo
- Trieve Librarian — Chrome extension for semantic bookmark search. Repo
- Titran — Esoteric programming language with custom interpreter. Repo
Latest Blog Posts
Your AI Coworker Should Be Boring (RPA Was Right All Along)
Hello World!
First ever blog post.