Selected Works
Research and engineering work across quantitative finance, graph-based modelling, deep reinforcement learning, computer vision, and full-stack systems.
Interpretable spatiotemporal Graph Neural Network with integrated causal discovery, modelling relationships between climate variables and regional disease outbreak patterns across Pakistan. Validated against epidemiological ground truth. Accepted & presented at ICHR-26.
Reduced post-Iftar hyperglycaemia by 35 mg/dL; achieved zero severe hypoglycaemia versus 4.5% failure rate in standard protocols; outperformed clinical sliding-scale protocols in 92% of simulated scenarios. Optimised for Ramadan fasting and high-glycaemic South Asian diets.
Multi-dimensional statistical framework for benchmarking large language model factuality, consistency, and reliability. Research Assistantship, IM|Sciences. Co-author. Manuscript in preparation for journal submission.
Quantitative asset-allocation algorithm using Bayesian optimisation to maximise risk-adjusted returns. Implements acquisition functions (Expected Improvement, Upper Confidence Bound) for efficient hyperparameter search. Robust out-of-sample performance with reduced overfitting versus baseline strategies.
Full-stack AI timetable platform serving 1,200+ students with intelligent routing between static schedule data and LLM-based responses. Serverless backend proxy for API security, model failover, streaming; admin analytics dashboard; offline-capable PWA with push notifications. GitHub →
Production-grade MT5 trading system with real-time market-data pipelines, statistical feature engineering, basket take-profit logic, regime-aware signal integration, and Dockerised CI/CD deployment on AWS.
Hybrid inference backend routing queries between local and cloud-hosted LLMs via a secure serverless proxy with adaptive cost-quality optimisation. AI study companion using the Socratic method to foster academic integrity and independent learning.
Production-grade object detection achieving 0.801 mAP for waste classification at 10 ms inference latency. Optimised pipeline using TensorRT and model quantisation; deployed via Streamlit for real-time smart recycling.
Two-tier conversational AI combining rule-based Rasa intents with GPT-3.5 fallback generation. Achieved 90% response accuracy and reduced average resolution time by 40% through contextual dialog management.
AI-powered platform for entrepreneurs to analyse market data, identify opportunities, track trends, evaluate competitors, forecast demand, and assess risks by combining real-time market data with AI analysis.
Award-winning multimodal deepfake detection system combining audio spectral features and visual artifact analysis to identify AI-generated misinformation. Media-literacy platform recognised at the UNESCO Hackathon for technical innovation.
Crowdsourced security platform connecting ethical hackers with organisations. Features vulnerability reporting, automated triage, and secure communication protocols. Incubated at the National Incubation Center.
Predictive models identifying customers likely to churn, enabling proactive retention strategies. End-to-end pipeline from feature engineering to model selection.
Streamlit web application performing real-time face detection and emotion classification using computer vision and deep learning models for server deployment.
AI-powered platform analysing satellite imagery, weather data, and social media to identify disaster-hit areas and predict damage extent for emergency response coordination.
Interactive system translating voice into musical and visual art, capturing emotions through stories and mapping them into visual and auditory forms of storytelling.
Complete CI/CD setup demonstrating automated testing, building, and deployment. Includes a REST API providing basic mathematical operations as the deployment target.
Reference materials for Digital Image Processing alongside an OCR project focused on Visual Pattern Recognition. Test images, key DIP concepts, and MATLAB implementations.
AtomCamp
Intensive training in statistical modelling, ML deployment, computer vision (YOLO), NLP, and LLM fine-tuning.
Aspire Institute
40-hour leadership development program focused on critical thinking, communication, and global impact. Selected for demonstrated academic excellence and leadership potential.
Supported initiatives for sustainable development and global problem-solving.
Engaged in discussions and workshops focused on youth-led climate action and initiatives aligned with the UNFCCC.
Completed coursework on Climate Change and Human Rights, learning advocacy for people affected by climate issues.