PyTorch · Transformers · NumPy · TorchVision · Diffusion Models · GANs · VAEs

A deep dive into generative modeling.

• Generative Models: β-VAE, GAN, CycleGAN, DCGAN, WGAN-GP (image generation and latent-space interpolation)
• Diffusion Pipelines: DDPM, DDIM, latent diffusion with conditional U-Nets and classifier-free guidance
• Efficient Vision: ViT with top-k sparse attention + convolutional stem; pruning + knowledge distillation for compression

React · Node.js · Express · MongoDB · Socket.IO

Full-stack browser-based Ludo with live multiplayer; synchronized game state, server-side move validation, and persistent game history across sessions. Built with the full MERN stack.

Android (Java) · Firebase · Figma

Led a 5-person team to build an MVC-patterned Android app for visitor management and credential verification at a university gate. Built with 2-week Agile sprints, UML diagrams, and real SWE practices.

Python · Scikit-Learn · Random Forests · GMMs · PCA

Built in a team of 3. Used Random Forests for feature selection to identify the most predictive audio attributes, then clustered users into taste profiles with Gaussian Mixture Models. Applied PCA to reduce and visualize the cluster space.

Python · NumPy · Custom CNN

Trained a deep CNN on CIFAR-10 from a fairly bare-metal setup — custom SGD with momentum and learning rate decay, no high-level wrappers carrying the weight. Hit 82% accuracy.

Python · NumPy · Scikit-Learn

Built a multivariate polynomial regression model on time-series stock data to forecast short-term price movements. Evaluated with cross-validation and RMSE.

Python · NumPy

Implemented a Support Vector Machine from scratch - linear, polynomial, and RBF kernels, no Scikit-Learn SVM. Applied to academic performance data to flag at-risk students with high precision and recall.

C# · .NET · MSSQL · Factory Pattern

Terminal-based full-stack app built in a team of 4 for managing dramas and drama societies across Lahore. Connected to a live MSSQL database, followed the Factory design pattern, built with the .NET framework.

• 🧪 Researching Federated Learning with LoRA + LTH-sparse adapters on Llama-3.2-1B @ LUMS
• 📖 Going deep on diffusion model theory and efficient inference
• 🛠️ Building and breaking things with the MERN stack