AI Developer - Punjab University College of Information and Technology
Final Year Project | ASD Support System — Prediction, Therapeutic Games & AI Assistant. Conducted research and development for early detection of Autism Spectrum Disorder (ASD) in young children by analyzing medical and behavioral features using machine learning and deep learning techniques.
- Developed two interactive therapeutic games based on research insights: Autism Counting Game — supports cognitive skill development and Autism Animal Matching Game — supports social skill development in autistic children
- Built ARIA (Autism Response & Interaction Assistant), an AI-powered support system designed for parents, therapists, teachers, and autistic children
- Implemented RAG Pipeline — citation-backed responses from uploaded ASD-related documents
- Integrated Real-time Emotion Detection — enables empathetic response adaptation
- Developed Multi-turn Memory — maintains conversational context across sessions
- Created Role-Specific Conversational Modes — tailored interactions per user type
- Built Personalized Routine Generator — custom daily routines for autistic users
- Implemented Social Skills Trainer — interactive skill-building module
- Developed Progress Tracking Dashboard — monitors user development over time
- Deployed using Groq API (Llama3) and HuggingFace embeddings for cost-efficient, high-performance inference
Data Science Engineer
FIFA World Cup 2026 Tournament Simulator. Built an end-to-end FIFA World Cup 2026 simulator featuring a 6-layer pipeline spanning data ingestion to ML-based match and bracket prediction.
- Developed 4 ML models from scratch in pure Python (no ML libraries): Gradient Boosted Trees Classifier — 83% accuracy, +32% improvement over baseline
- Trained Poisson & Negative Binomial Goal Models — via MLE with 10% MAE improvement
- Implemented 2-Layer MLP Neural Network — with manual backpropagation from scratch
- Designed a dynamic Elo Rating System with weighted K-factors and goal-difference adjustments for real-time team strength tracking
- Implemented a backtracking algorithm for valid FIFA bracket construction adhering to tournament constraints
- Incorporated Monte Carlo Simulations for match outcome variability and uncertainty modeling across tournament stages
- Engineered feature pipelines using FIFA Rankings, StatsBomb, and Transfermarkt data for robust team strength estimation
- Conducted WC 2022 backtesting and model validation to ensure predictive reliability and robustness
