Research Interests
My research interests include probabilistic modeling, Bayesian optimization, LLM reliability and hallucination analysis, quantitative finance, real-time computer vision, evolutionary computation, and AI governance.
My long-term research vision is to develop reliable, interpretable AI systems that bridge the gap between theoretical stochastic models and production-grade deployments—enabling trustworthy decision-making under uncertainty across domains from generative AI to algorithmic trading.
Probabilistic Modeling
Bayesian optimization, stochastic processes, and uncertainty quantification in machine learning systems. Developing methods that provide calibrated confidence estimates alongside predictions.
LLM Reliability
Empirical evaluation of hallucination behaviour, prompt sensitivity analysis, and reliability metrics for generative AI. Building statistical frameworks to quantify and mitigate model failures.
Quantitative Finance
Regime-aware portfolio optimization, cointegration-based trading strategies, and swarm intelligence heuristics. Applying optimization theory to financial decision-making under uncertainty.
Computer Vision
Real-time object detection, low-latency inference pipelines, and performance optimization. Engineering vision systems that meet strict latency and accuracy requirements.
Evolutionary Computation
Genetic algorithms, swarm optimization, and metaheuristics for combinatorial problems. Designing efficient search strategies for complex constraint satisfaction.
AI Governance
Policy constraints and governance structures for safe deployment of generative AI. Connecting model-level reliability metrics to system-level ethical guidelines.
Selected Projects
Bayesian Optimization for Basket Trading
Quantitative asset allocation algorithm using Bayesian Optimization to maximize out-of-sample returns. Implements acquisition functions (Expected Improvement, Upper Confidence Bound) for efficient hyperparameter search. Python, NumPy, SciPy.
Provenance — UNESCO Hackathon Winner
Award-winning deepfake detection system for digital media verification. Multimodal analysis combining audio spectral features and visual artifact detection. JavaScript, NLP, Deep Learning.
WasteVision
Production-grade YOLOv5 waste classification achieving 0.801 mAP with 10ms inference latency. Deployed via Streamlit for real-time smart recycling. PyTorch, YOLOv5, Computer Vision.
Hybrid AI Chatbot
Two-tier conversational AI combining Rasa intents with GPT-3.5 fallback. 90% response accuracy, 40% reduction in resolution time. Rasa, GPT-3.5, FastAPI.
Market Intelligence Platform
AI-powered market analysis for entrepreneurs: trend tracking, competitor evaluation, demand forecasting, risk assessment. TypeScript, AI/ML.
CyberSafe-PK
Crowdsourced security platform connecting ethical hackers with organizations. Incubated at National Incubation Center. Django, PostgreSQL, Docker.
Technical Expertise
Languages & Frameworks
Python, JavaScript, SQL, NoSQL, Django, FastAPI, React, TypeScript
Machine Learning
TensorFlow, PyTorch, Scikit-learn, Hugging Face Transformers, YOLOv5/v8, NLP, GPT-3.5, Rasa
Data Engineering
NumPy, Pandas, SciPy, PostgreSQL, MongoDB, SQL Server, Oracle, ETL Pipelines
Cloud & DevOps
AWS, GCP, Azure, Docker, GitHub Actions, CI/CD, Microservices Architecture
Quantitative Methods
Bayesian Optimization, Genetic Algorithms, Stochastic Modeling, Statistical Inference
Visualization
Power BI, Matplotlib, Seaborn, Plotly, React Charts, Custom Dashboards