Sudarshan
Mondal.

An engineer who reads code closely — and a researcher who wants machines to do the same.

I studied Computer Science & Engineering at North Western University, where my thesis taught a convolutional network to read Bengali handwritten characters — my first taste of building a system that has to understand messy, human input. That work was later published at ICICCT 2020 (Springer LNCS).

Can a system understand a codebase well enough to fix it? That question is what pulls me from engineering toward research.

Since then I’ve worked across the stack in production teams — a distributed logging pipeline handling a million entries a day, a secure file-serving platform on AWS, high-throughput preprocessing services, and Android apps with real users. Most of that work is the slow, careful business of understanding code that already exists.

My research interest is automated program repair and agentic debugging: LLM-driven agents that localise a fault, reason about intent, and propose a patch that actually passes the tests. It sits exactly where my two threads meet — the rigour of academic evaluation on one side, the messy reality of production codebases on the other.