Software Engineer, AI Systems and Full-Stack
Computer Science and Engineering at BUET. I build systems where a language model has to be right, not just fluent.
The work splits about evenly. Half of it is models measured against somebody else's held-out test set: Bangla speech recognition on hours of overlapping conversation for DL Sprint at BUET CSE Fest, and a Bengali hallucination detector that had to decide whether a fluent answer was actually true, which reached third on the leaderboard at 0.922 F1.
The other half is products, usually built to a deadline and usually with a team. A screening tool for Parkinson's that explains every score it gives. A thermal inspection app that binds each reading to the panel it came from. A platform that pays people to clean up waste only once the cleanup is verified.
Both halves want the same thing from me, which is why I keep doing both. Read more about me, look through my resume or projects, browse my writing, or get in touch.
