I’m all about blending AI with real-world impact from medical imaging and federated learning to AI-driven judiciary apps and system-level OS tuning. Whether it’s training CNNs for explainability, working at Linux kernels, or exploring federated architectures, I love building solutions that are scalable, ethical, and future-ready.

• Adaptive Feedforward MLP Driven Anomaly Recognition Framework using the LSNM 2024 Dataset - This repository contains an optimized Feedforward Multi-Layer Perceptron (MLP)–based Intrusion Detection System (IDS) built using PyTorch and evaluated on the LSNM 2024 dataset, a modern large-scale benchmark for network anomaly detection.
• Federated Learning with Adaptive Optimization – Personalized FL achieving 97.92% client accuracy.
• PathMNIST-XAI: Explainable CNN for Medical Imaging - Lightweight Explainable CNN for PathMNIST medical imaging, achieving 91%+ accuracy with Integrated Gradients, SQLite attribution storage, and scalable PyTorch deployment.
• Performance Analysis of models using Microsoft Safety Incident Prediction - A complete comparative analysis of Machine Learning and Deep Learning models on the Microsoft Security Incident Prediction (GUIDE) dataset. This project evaluates SVM, Random Forest, XGBoost, CNN, and ResNet on 2M training and 0.5M testing samples of high-dimensional security telemetry.
• AgeWell Admin Portal - A lightweight Admin Portal, using Flask + SQLite3 web app for small NGOs and orphanages to securely record and track visitor entries, donations, managing expense, track events and hold health records and personal details of staff, abandoned kids and elders all at one place.
• Recipe Management Website - A Flask + MySQL web app to create, manage, and share recipes with image uploads, allergen tracking, and vegetarian/non-vegetarian categorization.
• Public Pulse – AI-powered judiciary app for transparency & trust in automated decisions.
• Face Anonymization – YOLOv8 model trained for detecting face which is useful for blurring faces in images to protect privacy, which possess threat and privacy concerns of individuals.

• Stop Memorizing Loss Functions in ML - Here’s How to Actually Understand Them - This article breaks down the most important loss functions in machine learning. Instead of dumping formulas, I’ll show you where each one belongs, a mnemonic to remember it, and insider commentary you won’t usually find in textbooks.
• How Federated Learning Solves the Accuracy Gap in Heterogeneous Data - I explored Federated Learning (FL) to address a common challenge: improving model performance across clients with heterogeneous data without sharing raw datasets.

• IEEE CommBattle Ideathon Winner (2024) – Built AI-driven judiciary prototype and won a crash prize.
• C-DAC Intern – Kernel tuning for BOSS OS to improve latency & CPU scheduling.
• International Delegate (Malaysia) – Tech diplomacy + global exposure.

Languages → Python, C, C++, SQL, Bash
AI/ML Lib → PyTorch, TensorFlow, Scikit-learn, OpenCV, Captum, NumPy, Matplotlib, Keras, Pandas, SciPy, Seaborn, XGBoost Systems & OS → Linux (Kernel, Shell Scripting, Memory Mgmt), SQLite, Git

• Federated Learning & adaptive AI for fairness
• **AI for Healthcare & Judiciary **
• Open Source & HPC AI integration

🔹 LinkedIn
🔹 LeetCode
🔹 Medium

“Its all about one nice sip of tea and a tiny commit everyday.”