Final-year BS Data Science student at The Islamia University of Bahawalpur, building end-to-end AI systems — from model to deployment — with a focus on real impact in agriculture, public health, and disaster resilience across Pakistan.

CropGuard AI — Live crop disease detection platform for Pakistani smallholder farmers.

• 6 deep learning models (EfficientNetB0), 94.3% avg. accuracy across 51 diseases, 6 crops
• Trained on 45,000+ real field images — rejected lab-condition datasets for real Pakistani field conditions
• Bilingual Urdu/English AI advisory via Gemini 2.5 Flash
• Stack: TensorFlow, FastAPI, React, Docker, Hugging Face Spaces

Mandi Price Forecasting — 7-day-ahead wholesale crop price forecasting for Punjab, Pakistan.

• Playwright-scraped real government market data
• LightGBM benchmarked against a naive baseline
• Live Streamlit demo
• Stack: LightGBM, Playwright, Streamlit

Flood Risk Prediction — Geospatial ML pipeline predicting flood risk across Punjab & KPK, Pakistan.

• Satellite + weather data (Sentinel-2, SRTM DEM, CHIRPS rainfall) via Google Earth Engine
• XGBoost classifier — F1 0.52, ROC-AUC 0.912, validated against real 2022 flood extent data
• Interactive dual risk-map (predicted vs. actual) + 4-tab Streamlit app
• Stack: XGBoost, GeoPandas, Google Earth Engine, Streamlit, Folium

Heart Disease Prediction — Heart disease risk classifier with a full model comparison.

• 5-model comparison; tuned Random Forest — 0.96 AUC
• Streamlit demo app
• Stack: Scikit-learn, Random Forest, Streamlit

Used Car Price Predictor — Fair-price predictor trained on real scraped Pakistani market data.

• Scraped ~3,900 live PakWheels listings; engineered Pakistan-specific features (import/assembly status, registration province)
• Tuned XGBoost — 9.9% MAPE, with honest per-segment error analysis
• Stack: Scikit-learn, XGBoost, BeautifulSoup, Streamlit

Olist E-Commerce Analysis — Do late deliveries lower review scores, and can repeat buyers be predicted?

• SQL analysis in PostgreSQL, hypothesis testing with effect sizes and bootstrap CIs
• Time-aware ML pipeline for repeat-purchase prediction, explained with SHAP
• Power BI dashboard
• Stack: PostgreSQL, Python, XGBoost, imbalanced-learn, SHAP, Power BI

CortexIQ — AI-powered data analytics platform.

• Upload a dataset, chat in plain English, get auto-generated Python analysis
• Automated ML model comparison + branded PDF reports
• Stack: FastAPI, Next.js, Groq, Scikit-learn, XGBoost

Yacht Hydrodynamics — Resistance Prediction — Residuary resistance prediction on the UCI Yacht Hydrodynamics dataset.

• 7-model comparison, physics-informed feature engineering
• Honest write-up of where hyperparameter tuning did (and didn't) help
• Stack: XGBoost, Optuna, SHAP

Languages:

ML / Deep Learning:

Data & Analytics:

AI Engineering:

Tools & Deployment: