D

Developed a web-based Learning Management System (LMS) for a preschool using React (Vite) and a self-hosted Supabase backend. The system was designed to digitalize the school’s workflow, transforming traditional paper-based processes into an integrated digital platform. Key features include attendance tracking, announcements, report management, and user role management. I also handled the deployment and maintenance of the system using Docker on a self-hosted infrastructure, ensuring reliability, scalability, and consistent system performance.

I designed and managed a hybrid cloud environment combining on-premise servers with Oracle Cloud VPS to create a flexible and scalable infrastructure. I established secure private networking between nodes using WireGuard VPN, enabling safe and efficient communication across the system. I deployed and managed containerized applications using Docker and Podman, and configured Nginx as a reverse proxy to handle secure public access and domain routing. Additionally, I implemented centralized storage using TrueNAS with RAIDZ1, utilizing NFS and SAMBA for file sharing. To ensure secure connections, I automated SSL certificate renewal using Certbot and cron jobs. I also built a self-hosted Platform-as-a-Service (PaaS) using Coolify for streamlined application deployment, and deployed production-ready applications using Next.js, FastAPI, and PostgreSQL.

I developed a CNN-based image classification model using PyTorch to accurately identify marine fish species from images. To make the model accessible, I built a Streamlit web application that allows users to upload images and receive top-k predictions with confidence scores. I also implemented data preprocessing pipelines and evaluation metrics to optimize model performance and ensure reliable predictions.

Solar Intelligence Solar Intelligence is an AI-powered MVP designed to address the challenges of solar energy intermittency in tropical climates. By leveraging advanced machine learning, the platform mitigates the impact of rapid cloud movement and heavy rainfall on power grid stability. Core Features & Technology - High-Precision Forecasting: Utilizes an XGBoost ensemble learning model to provide 24-hour power output predictions, significantly reducing RMSE (Root Mean Square Error) compared to traditional methods. - Microservices Architecture: Built with a React.js frontend for data visualization and a FastAPI backend for high-performance processing, supported by SQLite for telemetry logging. - End-to-End AI Lifecycle: Integrates the full pipeline from feature extraction and cross-validation to automated deployment via Docker containerization.