CS student · AI, machine learning and computer vision
I chase questions, fall down rabbit holes, and build my way out.
out of 4.00
Dean's List four times, Rector's List twice
a job in AI, or grad school
What I work on
Getting a model to find and make sense of what is actually in a picture.
This is the area my final year project sits in. I am building a framework that uses a language model to help find several objects at once in a scene, from either a written description or an example image, for cameras that know where they are pointing. It is due in 2027. Before that I rebuilt a published model that sharpens Sentinel-2 satellite images, scaled down to run on my own machine. Going through someone else's design line by line taught me far more than reading about it ever did.
Building things with them, and checking what they are really doing.
I made a tool that shows what a language model is doing as it writes, one word at a time. It shows how confident the model is, where it is paying attention, and how likely it is to be making something up. Then I tested whether the model's own confidence could be trusted. It could not. When it said how sure it was, it was off by about 32 percent. Its own internal numbers were off by about 15 percent. I like working across all of this, from building with these models to working out why they behave the way they do.
Starting from the maths, not from someone else's library.
Neural networks, CNNs, ensembles and reinforcement learning. Some of it came from my coursework, some from Andrew Ng's specialisation, and a lot of it from building things until they finally worked. I would rather write a method myself and watch it break than trust a library to be right. This is the layer sitting underneath everything else on this page.
Selected work
012027
Final year project
My final year project, which I am working on now. It is a framework that uses a language model to help find several objects at once in a scene. You can describe what you are looking for in words or show it a picture, and it works with cameras that know where they are pointing.
Still deciding on parts of the stack.
022026
LLM tooling
A small web app that opens up how a language model thinks. It shows you, one word at a time, how confident the model is, where it is looking, and how likely it is to be making things up.
032025
Paper implementation
A model that sharpens blurry Sentinel-2 satellite images. I built it by following a published paper and scaled it down so it would run on my own machine, so I could understand exactly why the design works.
042025
Game / agent behaviour
A murder mystery game where the characters decide things for themselves. They patrol, chase and attack on their own, and they talk to you using a language model.
052025
Full stack platform
A booking and scheduling system for an airline, with an admin side for staff. It runs on a proper database behind a Next.js frontend.
062024
Desktop app
A desktop app that helps FAST students search old exam papers. It has a chatbot built in, and Supabase handles logins and file storage.
072024
Desktop app
A desktop social app with friend requests, messaging and user profiles.
About
I am in my final year of Computer Science at FAST-NUCES Lahore. My CGPA is 3.86. I have made the Dean's List four times and the Rector's List twice.
Most of what I build starts the same way. I want to see inside the thing. That is how I ended up writing a debugger that reads a language model's confidence one word at a time. It is why I rebuilt an image model from a research paper instead of just reading it. It is why I put decision trees into game characters, so I could watch them choose what to do.
I am heading into AI and machine learning work. I am keeping my options open between engineering jobs and graduate study while I finish my degree. Computer vision, language models and deep learning are the areas I want to keep building in.
FAST-NUCES Lahore
Clear Wave Information Technologies
OSOL Technologies
DeepLearning.AI, taught by Andrew Ng
AI / ML
Languages
Web
Tools