About
I build machine-learning models and the products around them.
I work across computer vision, efficient edge AI, full-stack development, and chatbots.
GitHub profileComputer vision
Segmentation models and neural networks built from scratch.
Efficient AI
Optimizing models for fast CPU and edge-device inference.
Full-stack products
Web apps built from the interface through the backend.
Chatbots & applied AI
Conversational tools that make AI useful in real products.
Skills
Programming & Tools
Deep Learning
Computer Vision
Edge Deployment
Full-Stack & AI Applications
Projects
Image Segmentation with U-Net
Python, PyTorchBuilt and trained a U-Net model predicting masks for objects. Added image augmentation for higher accuracy, achieving a validation Dice score of 0.85 and IoU of 0.76.
CNN Study on CIFAR
Python, PyTorchImproved a CNN from 75% to 87% accuracy. Implemented dropout, BatchNorm with GELU, learning rate schedulers, and fixed data augmentation leaks between training and evaluation.
Indus Dragon: Chess Engine
C++, PyTorch, AVX2Trained a custom MLP on 25M self-play positions. Deployed for ultra-fast CPU inference using AVX2 & INT8 vector operations. Achieved 3M+ nodes/second and 2200+ ELO.
Neural Networks from Scratch
Python, NumPyBuilt MLPs, CNNs, RNNs, LSTMs, GRUs, and a full GPT-1 style Transformer entirely without ML libraries. Implemented core optimizers including Adam, SGD, and RMSprop.
Research & Writing
Ideas I plan to explore and write about
Practical research notes connecting machine-learning concepts with systems that work in the real world.
Experience & Education
Experience
Contributor
Chesskit & Infinitunes
Education
BS Artificial Intelligence
PAF-IAST, Pakistan • Exp. 2028Coursework: Machine Learning, Deep Learning, Cloud Computing, Algorithms and Data Structures.