I'm a final-year undergraduate in Artificial Intelligence and Data Science, focused on building ML systems that go from research-grade modeling to production deployment. My work spans reinforcement learning on graph-structured data, MLOps pipelines, and applied deep learning — with an emphasis on rigorous evaluation, not just working code.
- 🔬 Currently exploring reinforcement learning, graph neural networks, and LLM-based systems
- 🛠️ Building end-to-end ML pipelines — from data ingestion to deployment
- 📫 Reach me at: gnikhilsai7@gmail.com
- 🤝 Open to research collaborations, ML internships, and interesting technical discussions
A node-scoring GCN-PPO framework for budget-constrained epidemic mitigation on temporal contact networks, modeling quarantine and vaccination as separate learned policy heads within a custom 8-state SEIQVR Gymnasium environment.
- Achieved 47–52% infection reduction vs. uncontrolled spread across two real-world SocioPatterns datasets (hospital N=75, school N=242)
- Validated results with Mann-Whitney U tests (p < 10⁻⁵) and Cohen's d > 0.8
- Outperformed DQN baseline by 21% and single-intervention ablations by 37–41%
- Two-phase training: behavioral cloning warm-start + PPO fine-tuning, validated across four ablation studies
Stack: Python, PyTorch, Gymnasium, NetworkX, NumPy, SciPy
An end-to-end MLOps pipeline for phishing URL classification using XGBoost, achieving 98.78% F1, 98.56% precision, and 98.99% recall on 11K+ samples with 30 network-level features.
- 6-stage modular pipeline (Ingestion → Validation → Transformation → Training → Evaluation → Pusher) with KS-test drift detection and automated model acceptance gating
- Experiment tracking via MLflow across 11+ training runs
- Weekly retraining and batch inference orchestrated via Apache Airflow, with AWS S3 artifact sync
- FastAPI service for training/inference, containerized with Docker, deployed to AWS EC2 via GitHub Actions CI/CD
Stack: Python, XGBoost, FastAPI, MLflow, Airflow, Docker, AWS (S3, ECR, EC2), MongoDB Atlas
Reinforcement Learning · Graph Neural Networks · MLOps · Deep Learning · LLM Systems & RAG · Applied Statistics