Computer Vision · Deep Learning · Edge AI · Autonomous Systems
📍 Lahore, Pakistan · 🎓 BS Computer Science, University of Management and Technology
Most student AI portfolios stop at a notebook with a validation curve. Mine doesn't — every project below either exports to an edge target, runs in real time, or gets scored against a real leaderboard.
- Healthcare cost prediction pipeline — multi-table preprocessing across claims, CPT/DRG/ICD codes, and member demographics, feeding an XGBoost + LightGBM + CatBoost ensemble. 0.94 ROC-AUC identifying high-cost members from ~45K stratified records.
- AI-driven urban traffic optimization — YOLOv8 detection → Transformer-based flow forecasting → VAE anomaly detection → a custom DQN controller (trained from scratch) closing the loop on signal timing.
- Gender-Identifier (YOLOv5) — real-time classifier on a self-built 2.6K-image dataset, >90% accuracy, served live through a Flask webcam stream.
- 🧠 Scoping my Final Year Project
- 📘 Working through Gerons Hands-On ML
- 🚁 Off-script but deliberate: teaching myself electronics → STM32/FreeRTOS → control theory
AI / ML
Deep Learning Computer Vision Digital Image Processing Vision Transformers YOLOv5 / YOLOv8 Reinforcement Learning Generative AI Anomaly Detection
Frameworks
Gym PyTorch TensorFlow / Keras Scikit-learn XGBoost LightGBM CatBoost OpenCV Hugging Face Optuna
Systems
CUDA C++ Python FastAPI Flask Docker Linux SQL
Building toward (drone-engineering track)
STM32 FreeRTOS PID Control Sensor Fusion
| Repo | What's actually in it | Stack |
|---|---|---|
| MachineLearning | 244+ commit hands-on ML/DL log — healthcare cost pipeline, residual networks, custom training loops, PCA + GPU-accelerated boosting, clustering, Flask/Streamlit deployment | PyTorch, TensorFlow, Scikit-learn |
| CUDA_Programming | Parallel kernels from scratch — vector ops, Euclidean distance, primality testing, memory management, CPU vs. GPU benchmarking | CUDA, C++ |
| Gender-Identifier-YOLOv5 | Real-time classifier, >90% accuracy, custom dataset, live webcam inference | YOLOv5, OpenCV, Flask |
| AI-Driven Urban Traffic Optimization | Detection → forecasting → anomaly detection → RL control, end to end | YOLOv8, Transformers, RL, PyTorch |
| Healthcare Cost Prediction | Ensemble pipeline, 0.94 ROC-AUC on high-cost member identification | XGBoost, LightGBM, CatBoost |
| NumJig | DSA practice repo (147 commits) — includes a sliding-tile puzzle game built on stacks, linked lists, graphs, and hashing to exercise the structures directly rather than just study them | C++, DSA |
| OOP-in-C-Plus-Plus | A hotel management system covering classes/objects, multi-level inheritance, static & dynamic polymorphism, encapsulation, abstraction, and friend functions — file handling via fstream, console animation via windows.h |
C++, OOP |
| Programming_Fundamentals_in_C_Plus | Console-based contact book app from a 14-week Programming Fundamentals course — CRUD on contact records using pointers, dynamic memory, and fstream persistence |
C++ |
Open to AI/ML role and research collaborations — especially anything that involves making a model run somewhere it wasn't supposed to fit: a GPU cluster, an edge board, or a flight controller.



