A quick look at how I think about building AI systems — and the stack I reach for.
I'm Muhammad Hassan, an AI developer focused on turning research-grade models into things people can actually use — a chatbot that answers correctly, a vision model that runs in production, an API that doesn't fall over under load.
My work sits across machine learning and deep learning foundations, computer vision pipelines for detection and recognition tasks, and NLP / LLM systems — from fine-tuned classifiers to retrieval-augmented chat assistants. I care as much about inference latency and clean API design as I do about model accuracy.
Most projects start the same way: a real constraint (data, compute, latency, cost) that shapes every decision after it. I like that part — it's where engineering and modeling actually meet.
A few systems I've designed and shipped — each solving one concrete problem end to end.
A real-time object detection pipeline that scans retail shelf footage to flag out-of-stock and misplaced items as they happen — the dashed box is the one it just caught.
Upload one or more PDF/TXT documents and ask questions in plain language — the assistant retrieves the relevant passages and answers with sources cited.
Upload a scanned or digital invoice/contract as a PDF or image — the app reads it and pulls out key fields like dates, amounts, vendor, and invoice number into a clean table.
A WhatsApp-style assistant that takes restaurant reservation requests through natural conversation, then extracts a clean structured booking card — no missed DMs, no manual back-and-forth.
Turn on your webcam and ask what it sees — a chat assistant answers in real time, grounded strictly in live YOLOv8 detections instead of guessing at the scene.
Tell me what you're working on — I read every message and reply within a day or two.