I'm an ML Engineer and Researcher pursuing M.Tech in Computer Engineering at NIT Kurukshetra, specializing in Federated Learning, GPU-accelerated Computer Vision, and LLM fine-tuning.
My research focuses on Non-IID data challenges in Federated Learning, where I developed Clust-PSI-PFL — a novel clustering framework using Population Stability Index that reduces communication overhead by 40% and improves model accuracy by 15% on skewed datasets.
I'm currently extending this work with formal security mechanisms — integrating Differential Privacy and Byzantine-robust aggregation into the FL pipeline. I also build production-grade GenAI systems including RAG pipelines with Mistral-7B fine-tuning.
Outside research, I lead campus placement operations as TPO Coordinator at NIT Kurukshetra, coordinating 50+ companies and facilitating placements for 150+ students annually.
We propose a novel federated learning framework that leverages Population Stability Index (PSI) as a lightweight client clustering metric to address Non-IID data heterogeneity. By grouping clients with similar data distributions via K-means++ on PSI features, our method achieves 15% accuracy improvement on CIFAR-10/FEMNIST while reducing communication overhead by 40% compared to gradient-based clustering methods. Benchmarked against FedAvg, FedProx, and Per-FedAvg — outperforming all on Non-IID partitions.
Novel clustering-based FL system tackling Non-IID data challenges across distributed GPU-accelerated networks using Population Stability Index as a lightweight clustering metric.
End-to-end RAG pipeline with FAISS dense + BM25 sparse hybrid retrieval using Reciprocal Rank Fusion. Fine-tuned Mistral-7B with QLoRA for domain-specific question answering over research papers.
GPU-accelerated CNN achieving 98% accuracy across 30+ disease categories. Automated fertilizer recommendation module mapping detected diseases to specific treatments.