Hi, I'm Ashutosh Roy

It started in Faridabad, Haryana. 94 in 10th. 87 in 12th. Good enough to dream bigger. So I aimed for IIT — gave JEE, cleared Mains, hit a wall at Advanced. Took a drop year. That year was quiet and heavy, but it didn't break anything. If anything, it built a kind of stubbornness that never really left.

College began at CSVTU, Bhilai. The first year was all maths and statistics — no real coding yet. Then came C, then Python, then R somewhere in between. Nothing serious, just poking around. Until ChatGPT and Claude landed, and everything shifted. I remember thinking: I want to build things with this. So I started. Months of breaking pipelines, debugging at midnight, learning what "automation" actually means when you're doing it alone. Built a YouTube content pipeline from scratch. Published some videos. Most things broke before they worked.

Then 3rd year hit — and with it, that familiar campus panic: everyone needs an internship, now. Instead of applying the slow way, I built BulkyMail — a tool to send hundreds of personalized emails to professors and company HRs in minutes. Used it myself. Half my batchmates used it too. Some of them got internships because of it. That felt good in a way no grade ever did.

Around the same time I went deep into ML — algorithms, model fine-tuning, the actual bones of how these systems work. Something about it just clicked. And then the opportunity came: an internship at IIT Delhi, under Prof. Maya Ramanath. I was in a lab that worked on real problems — scanned archaeological manuscripts, decades of handwritten field notes with no structure and no search. We built the whole thing: OCR pipelines, benchmark comparisons, hybrid semantic retrieval. It became the SARCH system. I co-authored the paper with the IIT team. The day it went up on arXiv, I read my name on it three times just to make sure.

Final year brought a new obsession. During IIT, between the late nights and the research, I was also recording — teaching RAG on YouTube while actively building it. Then fell deep into audio and vision models — how they hear, how they see, how they generate. Got fixated on AI music generation, started reverse-engineering how models like Suno work, made AI music videos just for the fun of understanding them. That hands-on obsession with audio and vision models fed directly into what came next — a second research paper on speech emotion and stress detection, accepted at RECCAP 2026 and published in IEEE Xplore. The foundation was already there. The paper was just the proof.

Somewhere through all of this, MITOVOID AI started taking shape — a startup, still early, still being built one late night at a time. I play badminton when I need to reset. I share the journey on my YouTube channel and post tech discoveries as @atrexplains. The degree from CSVTU is almost done. But the learning — that never waited for a deadline.