nairadithya

adithyanair [dot] dev [at] gmail [dot] com| linkedin | github | twitter | now | cv

Hello, I’m Adithya Nair (@nairadithya). I build ML and robotics systems: world models, state estimation, and the low-level software around them.

I’m finishing a B.Tech in AI & Data Science at Amrita Bengaluru (graduating May 2027) and looking for ML / robotics roles.

What I’ve Worked On

1. rover control plane - turned a Dell OptiPlex into a computer ready for robotics: a FastAPI + rosbridge server exposing ROS 2 control, camera streaming, and boot-time device management.
2. learned world models for multi-modal sensor fusion - radar + EO/IR target tracking with learned dynamics, benchmarked against classical Kalman filters.
3. JEPA world model over longitudinal MRI - predicts the latent of a patient’s next MRI visit from prior scans; held-out test loss 0.0074.
4. RL post-training for small language models - GRPO and DAPO training sweeps on Qwen and Gemma for mathematical reasoning.
5. mentat - a multi-threaded OCaml 5 HTTP server; 180,000+ requests per second on a consumer laptop.

Community efforts

An interesting problem i like to think about is how i can build communities full of inquisitive and likeminded folks. I host two communities:

1. amrita.town - an indieweb webring in my college. It’s focused on personal websites and trying to encourage people to take advantage of the web to express themselves. This community of people has slowly morphed into becoming an unofficial student community that builds and maintains software for the benefit of students:
   • amrita timetable app - an android app that provides a simple widget to check your timetable. We crowdsource the timetables sent to students by their departments and standardize them into a registry of JSON files accessible at https://timetable-registry.amrita.town/
   • timetab.el - emacs plugin to create an org-mode file of a student’s time table.
2. atlas - a systems programming club and discussion forum. We conduct workshops and have daily discussion threads.
   • terminal velocity - a workshop focused on turning students into terminal wizards.
   • links archive, An archive of high quality links on the craft of programming and systems.

I am interested in learning

1. GPU Programming - I want to get my hands dirty with some real gpu hardware and learn cuda.
2. Robotics - My taste of robotics within the rover team wants me to dig deeper. I had a robotics course and i know some rudimentary embedded programming, part of me wants to dig deeper.
3. Frontier Model Architectures - implementing model architectures from papers.