
Things I've Built
Experiments, shipped products, and late-night rabbit holes that turned into something real. Every project taught me something new.

The Signal - AI Newsletter
An automated n8n pipeline that scrapes 10 AI/ML/tech RSS feeds and publishes a daily digest to GitHub Gist without manual intervention. Deployed on a self-managed DigitalOcean VM with a custom domain and SSL, running on a daily schedule. A Next.js frontend on Vercel presents the digest in a readable format.
- n8n
- Next.js
- Gist
- Vercel
- +2

Serendib - By Ceylon Explorer
An AI-powered conversational travel-planning avatar for Sri Lanka trip planning, built within a 24-hour buildathon for the "Beyond Presence" track at Cursor Colombo 24H Buildathon. Implemented an n8n-based workflow (self-hosted on DigitalOcean) that captured leads, triggered the conversational agent, and auto-sent trip summary emails to clients and organizers.
- Beyond Presence
- Agentic AI
- n8n
- Next.js
- +2

DelBuddy
DelBuddy is a lightweight, gamified desktop application that scans your machine for the junk you've been too bothered to deal with, and helps you delete it in a fun, interactive way.
- Electron
- React
- Node.js
- creative solution


Debater - AI by GDesh.dev
A multi-agent AI application in which two AI personas autonomously debate opposing sides of a topic. Designed the adversarial agent architecture and orchestration logic for structured, multi-turn debate and conclusion generation.
- Next.js
- LangChain
- Gemini API
- LLM
- +2
My-Vault - Password Manager & Secret Vault
A premium, zero-knowledge password manager and secret vault built with Next.js 16, React 19, and Supabase. Passwords, API keys, secure notes, and cards are encrypted entirely in the browser using AES-256-GCM via the Web Crypto API — the server only ever sees ciphertext. Features a dual password system (account + master), TOTP 2FA, a cryptographically random password generator, vault search/filter/favorites, clipboard auto-clear, and Supabase row-level security.
- Next.js
- React
- Supabase
- TypeScript
- +3

Memory Access Guard (MAG+)
A kernel-level memory protection and logging framework for xv6-riscv that intercepts illegal memory accesses via trap handling. Classifies fault regions (null dereference, unmapped memory, buffer overflows) and logs structured fault data in a circular kernel buffer. Exposes fault logs to user space through a custom system call (memlog) and ensures only the faulty process is terminated, keeping the rest of the system stable.
- C
- xv6-riscv
- Kernel Development
- Memory Protection

J'puraXtreme 2.0 - Official Website
Developed the web presence for J'puraXtreme 2.0, a two-round competitive programming event organized as part of IEEEXtreme preparation at the University of Sri Jayewardenepura. Built a Batman-themed event experience aligned with the competition's branding and promotional campaign. Supported participant communication and event visibility through a centralized web platform.
- Next.js
- Tailwind CSS
- Vercel
- Event Website

Blood Inventory Management System (BIMS)
A role-based blood inventory management system supporting donor and hospital workflows for blood donation and inventory coordination. Built with Laravel. Implements separate authentication and system functionality for donors and hospitals, enabling structured management of blood-related records and requests.
- Laravel
- PHP
- MySQL
- Web Development

Power BI Marketing Strategy Dashboard
An interactive Power BI dashboard to analyze streaming service performance and support a December 2025 marketing strategy. Designed KPI cards and time-based visualizations to track customer, engagement, and performance trends across key business metrics. Transformed analytical findings into actionable marketing insights through focused dashboard storytelling.
- Power BI
- Data Analytics
- Dashboard
- Visualization
Customer Churn Prediction Model
A machine learning pipeline to predict customer churn using the Telco Customer Churn dataset. Evaluated multiple classification models — Logistic Regression, Random Forest, KNN, and XGBoost — using Accuracy, Precision, Recall, F1-score, and ROC-AUC. Applied data preprocessing and model evaluation techniques to identify high-risk customers and support data-driven retention strategies.
- Python
- Machine Learning
- XGBoost
- Classification
- +1
Diabetes Risk Factors Analysis
Analyzed a large-scale diabetes health indicators dataset containing over 250,000 records and 22 health-related variables to investigate factors associated with diabetes risk. Performed exploratory data analysis and statistical testing, including frequency analysis and Chi-square tests and visualize key risk factors.
- SPSS
- Power BI
- Data Analytics
- Statistics
- +1