ML & AI Projects

A Qwen 2.5 7B Instruct model fine-tuned with QLoRA to speak natural Urdu: Urdu script, Roman Urdu, and Urdu/English code-mixing. The current version reaches a 79.5% pairwise win rate over the base model across two independent LLM judges on a 100-prompt evaluation set, recovers every regression its predecessor introduced, and adds RAG-aware training that makes retrieval safe on factual queries. Validated with a multi-judge harness (including a free Claude-Code-CLI judge) and shipped as a live Gradio + Modal demo. A first end-to-end fine-tune, built in public for about $60.

Machine learning system that predicts US PERM labor certification outcomes and explains why using SHAP. Features a 5-stage MLOps pipeline, GridSearchCV across 5 boosting models, threshold-tuned Gradient Boosting, and a FastAPI backend with per-prediction explainability.

A deep learning web application that detects and classifies facial expressions in images. Built with PyTorch and Flask, featuring a ResNet-18 model trained on the RAF-DB dataset achieving 80% accuracy.

Machine learning system for groundwater monitoring and prediction in Pakistan, built on 22 years of satellite data (2002-2024) from GRACE and GLDAS. Features time-series forecasting with Gradient Boosting (R²=0.89), interactive district-level maps, and a Flask API for predictions.

A teacher rating and review platform that lets students share structured feedback about educators across multiple dimensions, helping others make more informed decisions during course registration. Built with React, TypeScript, and Supabase.

Multi-criteria spatial optimization for EV charging station site selection in Lahore using weighted scoring algorithms, demographic analysis, and geospatial data to find the best locations for charging infrastructure.