01
Evidence over theater
A polished demo is useful, but I trust reproducible evaluations, failure cases, and clear limits more. That is why my project pages show the runs that failed as well as the ones that worked.
About
I like difficult systems and plain explanations.
I'm Sohail, an AI/ML engineer and M.S. Computer Science student at USC. I work on systems where model behavior has to meet real software constraints: autonomous code migration, multimodal medical-AI research, retrieval pipelines, and the interfaces that make those systems usable.
I started my career at IIFL building internal AI products for employees and support teams. That experience shaped how I think about applied AI: the model is only one part of the product. Reliability, permissions, observability, and a clear fallback matter just as much.
At Keck School of Medicine of USC, I now work across imaging, clinical data, multimodal learning, and retrieval-augmented VQA. Outside research, I keep building tools that let me test ideas end to end instead of stopping at a notebook.
How I work
The principles behind the projects.
02
Own the whole path
I am happiest moving between model behavior, backend boundaries, data, and the interface. The seams between those layers are usually where the most important problems hide.
03
Make complexity legible
Good engineering should leave the next person with a system they can understand. I care about observable workflows, honest naming, and documentation that explains decisions rather than restating code.
Beyond the Code
Stuff I care about when I'm not coding.
Marvel Universe
Favorite
Spider-Man
No matter how beaten, how outmatched, he gets back up. Every time. That kind of resilience is something I try to carry into my own life.
Favorite
Iron Man
Always has a backup plan. And a backup for the backup. If one thing goes south, there's already another plan ready. I try to think like that when I'm building systems.
Story-Driven Games
Simple Pleasures
Tap any to learn more