Chest X‑ray Pneumonia Detector
Built a chest X‑ray pneumonia detector on the Kaggle Chest X‑Ray Pneumonia dataset using a DenseNet‑121 backbone in PyTorch. Trained the model with lung‑focused preprocessing (central crops and random resizing) to reduce reliance on shortcuts such as text overlays and shoulder borders, and evaluated it on a 624‑image test set, achieving ~78% accuracy and AUC ≈ 0.89 while maintaining very high sensitivity to pneumonia cases. Integrated Grad‑CAM to generate heatmaps that highlight the lung regions driving each prediction, and deployed an interactive Gradio web app for exploring model outputs.
Handwritten Digit Recognition with CNN
Built and trained a Convolutional Neural Network on the MNIST dataset to classify handwritten digits (0–9). Implemented data normalization and Dropout regularization, and evaluated performance with accuracy and a confusion matrix. Includes training and inference scripts plus a simple demo for predicting from images.
Smart Recipe Finder
Developed a recipe app that searches and recommends recipes based on user-provided ingredients. Designed and implemented a recommendation algorithm using Python and Flask. Created a simple, responsive interface using HTML, CSS, and JavaScript and enhanced user interface using Vue.js
Hydroponic System
Led team of 5 in constructing a hydroponic system for indoor plant growth without natural sunlight.
Sentinel Stream
Distributed telemetry pipeline for real-time metrics ingestion and time-series persistence. Ingests high-throughput agent metrics via Redis Pub/Sub, persists to InfluxDB, and exposes a REST API — all containerized with Docker Compose. Designed for low-latency, concurrent data collection across distributed nodes.
Nexus Alloc
High-performance custom memory allocator in C/C++ exploring fragmentation reduction, memory alignment, and allocation strategies. Designed for deterministic, GC-free environments where predictable latency is critical.
SkillGap AI Coach
Full-stack AI-powered resume and job description analyzer. Runs a deterministic baseline scoring engine with an optional LLM mode for deeper analysis. Features persistent history, skill gap breakdowns, and actionable recommendations. Fully containerized with Docker Compose.
Image Editor Pro
Feature-rich desktop image editor built with PyQt6. Supports layers, filters, brightness/contrast controls, and full undo/redo via the command pattern. Clean MVC architecture with a polished dark-mode UI.
Transaction Monitor Service
Spring Boot REST API for transaction risk monitoring. Includes a full test suite with JUnit, CI pipeline via GitHub Actions, and clean layered architecture following industry best practices.
Risk Decision Dashboard
React/TypeScript dashboard for visualizing and exploring risk decision datasets. Features Redux Toolkit for state management, dynamic filtering, and a clean component-driven UI with dark mode support.