Hello, I'm Nithin Balamurugan
I'm a computer science student at Western University
Breaking into AI and research for Tiny Recursive Models (TRMs).
I'm a computer science student at Western University
Breaking into AI and research for Tiny Recursive Models (TRMs).
I'm a 3rd year student breaking away from just software development and into AI and new tech research. I'm currently working for a startup at Cognitome.ai, where I am working with webdev and hands on work with AI. I've co-authored our first paper "Tiny Recursive Models on ARC-AGI-1- Inductive Biases, Identity Conditioning, and Test-Time Compute".
I'm hoping to futher grow my knowledge and work expirience in a professional work environment and working towards a career in tech consulting and sales/product managment.
Built a full-stack AI finance chatbot using React, Node.js, and PostgreSQL to track expenses, manage budgets, and visualize spending. Integrated a Python-based AI backend (FastAPI, OpenAI API, Pandas) for natural-language queries and personalized insights; deployed with Docker on Heroku/Vercel.
Developed a high-accuracy PDF-to-CSV conversion algorithm using Python that extracts every table into separate CSV files, generates a full-document text file, and outputs structural metadata files detailing table outlines, dimensions, and formatting—robust across any document layout. Built a modular extraction pipeline using Docling, PyMuPDF, Pandas, and a 3-layer OCR stack.
Co-authored "Tiny Recursive Models on ARC-AGI-1: Inductive Biases, Identity Conditioning, and Test-Time Compute", analyzing the behavior of Tiny Recursive Models (TRMs) on the ARC-AGI-1 benchmark. Performed empirical ablations and efficiency analyses to isolate the impact of test-time compute, puzzle-identity conditioning, and recursion depth on model performance. Benchmarked TRMs against a QLoRA-fine-tuned LLaMA 3 8B baseline.
Developed a personal portfolio website to showcase projects in depth and share background and skills. Implemented using Next.js, React, TypeScript, server-side rendering (SSR), React Hooks, Tailwind CSS, PostCSS, responsive design, dark mode, animations, form handling, ESLint, Node.js, npm, and Next.js Google Fonts.
Built an automated self-watering plant system using Arduino to monitor soil moisture and trigger watering below a set threshold, enabling plant care during long absences. Integrated sensors and output modules (LCD, moisture sensor, LED, buzzer, water pump) using Java and Matlab to display moisture data and signal watering events.