# Projects ## MLOmics Project URL: https://github.com/deaneeth/multi-omics-cancer-subtype-classifier Skills: Research, PyTorch, XGBoost, scikit-learn, SHAP, Streamlit, TCGA, KEGG Summary: End-to-end multi-omics cancer subtype classifier using TCGA data, PyTorch fusion models, and explainable AI. MLOmics is a reproducible machine learning pipeline for cancer molecular subtype classification using multi-omics integration. It combines gene expression, microRNA, DNA methylation, and copy number variation data from 931 TCGA patients across breast cancer (BRCA) and colon cancer (COAD) cohorts. The project evaluates tree-based baselines and deep learning fusion models, including a pathway-aware fusion architecture that injects KEGG biological pathway knowledge into the mRNA encoder. It also includes explainability workflows and a Streamlit demo for prediction, model comparison, and lab-format data conversion. ## Telco Churn MLOps Pipeline Project URL: https://github.com/deaneeth/telco-churn-mlops-pipeline Skills: MLOps, Python, scikit-learn, PySpark, Kafka, Airflow, MLflow, Docker, Flask Summary: Production-grade MLOps pipeline for telecom churn prediction with Kafka streaming, Airflow orchestration, MLflow tracking, and Docker deployment. This project implements a production-grade MLOps pipeline for predicting telecom customer churn. It covers data ingestion, preprocessing, model training, experiment tracking, workflow orchestration, streaming inference, and REST API deployment. The pipeline is the final stage of a four-phase development series: - **Phase 1**: EDA, data exploration, and preprocessing - **Phase 2**: Model training, evaluation, and hyperparameter tuning - **Phase 3**: Production pipeline architecture with monitoring - **Phase 4**: Full MLOps integration - Kafka, Airflow, MLflow, PySpark, Docker, and API deployment ## Customer Churn Prediction Pipeline Project URL: https://github.com/deaneeth/churn-prediction-production-pipeline Skills: MLOps, Python, Pandas, NumPy, scikit-learn, XGBoost, LightGBM, FastAPI, Jupyter Notebook, Pydantic, Groq, Feature Engineering Summary: Three-repository churn prediction progression from EDA and preprocessing to model training, evaluation, and a production-oriented pipeline. Customer Churn Prediction Pipeline is a three-repository ML lifecycle project for churn prediction, split into data preparation, model training, and production-oriented pipeline structure. The flow starts with [churn-prediction-data-pipeline](https://github.com/deaneeth/churn-prediction-data-pipeline) for EDA, missing-value handling, outlier treatment, binning, encoding, scaling, and processed datasets. It then moves into [churn-prediction-model-training](https://github.com/deaneeth/churn-prediction-model-training), which covers baseline training, K-fold validation, multi-model comparison, hyperparameter tuning, threshold optimization, and saved artifacts. The main portfolio link points to [churn-prediction-production-pipeline](https://github.com/deaneeth/churn-prediction-production-pipeline), which turns the earlier notebook work into reusable source modules, pipelines, config.yaml, artifacts, Makefile automation, and a FastAPI-ready inference direction. This is separate from the Telco Churn MLOps Pipeline: this project shows the cleaner modular progression, while Telco is the heavier Kafka/Airflow/MLflow/PySpark/Docker implementation. Current status is honest: the production repo documents the data pipeline as functional, with training and streaming inference still being built out. ## TinyGPU Project URL: https://deaneeth.github.io/tiny-gpu/ Skills: Open Source, Python, GPU Architecture, SIMT, Parallel Computing, Visualization Summary: A lightweight Python GPU architecture simulator that visualizes parallel threads, registers, memory, and instruction execution. TinyGPU is a Python-based GPU architecture simulator built to make low-level GPU execution easier to understand. It models parallel threads, registers, memory, shared memory operations, synchronization, and instruction execution in a small educational simulator. The project is designed for students, engineers, and curious developers who want to understand how GPU-style parallel execution works without needing actual CUDA hardware. ## TinyTPU Project URL: https://tiny-tpu.vercel.app Skills: Open Source, SystemVerilog, WebAssembly, Astro, React, TypeScript Summary: A real synthesizable SystemVerilog 4×4 systolic array compiled to WebAssembly and visualized live in the browser. TinyTPU is a browser-based hardware visualization project that runs a real synthesizable SystemVerilog systolic array in the browser. The RTL is compiled to WebAssembly and visualized cycle-by-cycle so users can watch matrix multiplication move through processing elements. The project connects low-level computer architecture, hardware simulation, WebAssembly, and modern frontend visualization into one interactive demo. ## BrainUs AI Project URL: https://www.brainus.lk Skills: AI/ML, RAG, Python, FastAPI, Next.js, TypeScript, Vector Search, LLMs Summary: AI-powered educational platform for Sri Lankan O/L and A/L students with curriculum-grounded RAG retrieval and citation-aware answers. BrainUs AI is an AI-powered learning platform for Sri Lankan students, designed around curriculum-grounded support for O/L and A/L education. The platform focuses on making learning easier through intelligent retrieval, structured educational content, and accessible product design. Dineth contributes across AI/ML engineering, RAG architecture, backend API workflows, product design, and frontend implementation. The source code is private - this is a production platform actively serving students. ## Smart Beehive Monitor Project URL: https://github.com/deaneeth/smart-beehive-monitor Skills: IoT, ESP32, C++, Firebase, React, Next.js, TypeScript, Solar Power Summary: ESP32-based IoT system for real-time beehive monitoring with environmental sensors, weight, GPS, solar power, and Firebase dashboard. Smart Beehive Monitor is an ESP32-based IoT system for beekeepers that collects real-time data about hive environment, bee activity, hive weight, location, motion events, and power status, then syncs the data to Firebase for dashboard visualization. The project covers both the embedded hardware side (ESP32 firmware in C++) and a web dashboard (React/Next.js) for monitoring hive health and activity patterns remotely. The companion dashboard source is maintained separately at [smart-beehive-dashboard](https://github.com/deaneeth/smart-beehive-dashboard), while the hardware and firmware source lives in the main monitor repository. Related write-ups: [full-stack ESP32 IoT beehive case study](/blog/full-stack-esp32-iot-smart-beehive-monitor) and [AI beehive computer vision experiment](/blog/from-bees-to-bytes-building-ai-beehive). ## E-commerce Repeat Buyer Prediction Project URL: https://github.com/deaneeth/ecommerce-repeat-buyer-prediction Skills: PySpark, Databricks, Spark MLlib, Python, Big Data, RFM Segmentation Summary: Big data ML pipeline using PySpark and Databricks to predict repeat buyers from 20.7M clickstream events with Spark MLlib. This project predicts repeat buyers in e-commerce using Spark MLlib on 20,692,840 clickstream events from the REES46 dataset. It combines big data processing, model selection, leakage-safe temporal splitting, RFM segmentation, and campaign targeting analysis. The workflow benchmarks five ML model families, identifies the best performer (Gradient-Boosted Trees at AUC 0.9177), and translates model rankings into business-actionable campaign targeting strategy. ## Behavioral Biometric Authentication Project URL: https://github.com/deaneeth/behavioral-biometric-authentication Skills: Research, Deep Learning, MLP, MATLAB, Gait Recognition, Biometrics, Feature Engineering Summary: Deep learning-based behavioral biometric authentication using smartphone gait data with 3.43% EER and 92.68% accuracy. This project implements a behavioral biometric authentication system that identifies users from walking patterns using smartphone accelerometer and gyroscope data. Instead of passwords or fingerprints, the system explores passive authentication through gait recognition. The project uses multi-sensor fusion, feature extraction, MLP classification, cross-day validation, and systematic optimization to reduce Equal Error Rate from baseline to 3.43%. ## Enterprise Cinema Booking Platform Project URL: https://github.com/deaneeth/enterprise-cinema-booking-platform Skills: Full Stack, React, TypeScript, Node.js, MongoDB, WebSockets, Docker, RBAC Summary: Scalable multi-tenant cinema management platform with real-time seat reservation via WebSockets, RBAC, PayPal payments, QR tickets, and Docker deployment. Enterprise Cinema Booking Platform is a scalable, multi-tenant cinema management and ticket booking system built with the MERN stack, TypeScript, Docker, and WebSockets. The platform supports real-time seat reservations via Socket.IO, three-role RBAC, cinema administration with approval workflows, PayPal checkout with automatic refunds, QR code ticket generation, booking history, and separate management dashboards for users, cinema admins, and platform admins. ## FurnishView Project URL: https://github.com/deaneeth/furnish-view Skills: Full Stack, JavaScript, HTML5 Canvas, Three.js, WebGL, MVC, HCI, Computer Graphics, Usability Testing Summary: Zero-build browser room-design tool with a hand-coded 2D Canvas engine, Three.js 3D visualization, localStorage persistence, and HCI usability evaluation. FurnishView is a browser-based room design and furniture visualisation tool built for in-store furniture consultations. A designer can log in, create a room from real-world dimensions, place furniture on a 2D top-down canvas, generate an orbitable Three.js 3D scene, compare two saved layouts side by side, and export a customer-ready design document. The project was built for PUSL3122 HCI & Computer Graphics 2025-26. My role focused on the application architecture, 2D Canvas engine, Three.js rendering pipeline, and innovation features. The implementation uses strict MVC separation: models hold room/furniture/design state, views render DOM and bind UI events, controllers orchestrate business logic, graphics modules own all Canvas/WebGL rendering, and utilities handle storage, scaling, constants, and persistence. This is deliberately a zero-build web app. There is no React, no Vite, no backend, and no external runtime dependency requirement. The app runs from index.html with local libraries, persists users/sessions/designs through localStorage, and demonstrates the graphics and HCI work directly in the browser. The HCI side was not treated as decoration. The workflow includes formative think-aloud findings that changed the interface, plus summative testing with a SUS score of 78.75 and 100% task completion across the tested consultation tasks. ## Whisper Project URL: https://whisper-anonymous-gossip-web-application.vercel.app Skills: Full Stack, React, TypeScript, Gun.js, IndexedDB, Encryption, P2P Summary: Anonymous P2P gossip platform using cryptographic keys, AES-256 encryption, offline-first storage, and decentralized sync. Whisper is an anonymous P2P gossip web application that avoids traditional accounts and personal information. Users interact through cryptographic keys, encrypted data, local/offline storage, and decentralized synchronization via Gun.js. No server-side identity - the system uses cryptographic keys instead of account-based login, making user identity truly anonymous by design. ## BookBridge Project URL: https://bookbridge-dashboard.vercel.app Skills: Full Stack, Next.js, React, TypeScript, Firebase, Firestore, Cloudinary, Real-time Chat, Marketplace Summary: Secure used-book marketplace with Firebase auth, verified listings, advanced search, real-time buyer-seller chat, seller analytics, ratings, and admin moderation. BookBridge is a secure used-book marketplace built around the trust problems that usually make peer-to-peer book trading messy: unverifiable listings, weak search, unsafe off-platform communication, and no seller feedback loop. The platform uses Next.js, React, TypeScript, Tailwind CSS, Firebase Authentication, Firestore, Firebase Storage, and Cloudinary. Buyers can browse and filter books by book-specific metadata, inspect listing details, message sellers in real time, and use ratings and feedback to judge trust. Sellers get dashboard tooling for managing listings, tracking views, spotting trends, and adjusting prices. The admin side matters here. BookBridge includes moderation and listing verification workflows so the marketplace is not just a CRUD app with book cards. Admin-reviewed listings, protected routes, Firebase security rules, HTTPS, and role-based access are part of the core design. The README also documents a QA pass covering unit and integration testing with Jest/Mocha, Firebase auth/security checks, performance testing with a stated 1000+ concurrent-user scenario and sub-2s response target, usability testing, and acceptance testing against functional and non-functional requirements. ## AQI Monitoring System Project URL: https://github.com/deaneeth/aqi-monitoring-system Skills: Full Stack, Laravel, PHP, MySQL, Leaflet.js, JavaScript, Bootstrap, RBAC, GIS Summary: Laravel air-quality monitoring dashboard for Colombo with Leaflet maps, simulated AQI updates, threshold alerts, CSV import, role-based admin panels, and PHPUnit tests. AQI Monitoring System is a Laravel and MySQL web application for visualising air-quality conditions across the Colombo Metropolitan Area. It combines a public Leaflet.js map, admin dashboards, simulated AQI data generation, alerting, CSV import, and role-based access control into one monitoring workflow. The public side focuses on clarity: users can open the map, inspect colour-coded sensor markers, view detailed AQI readings, and understand the severity band immediately. The admin side adds the operational layer: sensor management, user management, alert configuration, CSV imports, diagnostics, and historical AQI charts. The system uses simulated AQI data for demonstration rather than live hardware. That was a deliberate academic constraint, so the simulation engine supports realistic patterns such as random fluctuations, sine-wave behaviour, peak-hour variation, configurable intervals, and AQI value ranges. Alerts are generated when AQI values exceed 200. The implementation is grounded in standard Laravel practices: PHP backend, MySQL schema for sensors, sensor data, users, and alerts, middleware-protected authentication, CSRF protection, REST-style JSON endpoints, Bootstrap-responsive Blade views, and PHPUnit unit/feature testing. ## Artiqo Mobile Marketplace Project URL: https://github.com/deaneeth/artiqo-mobile-marketplace Skills: Full Stack, Flutter, Dart, Firebase, Firestore, Firebase Auth, Cloudinary, Mobile UI, E-commerce, Localization Summary: Flutter mobile marketplace for Sri Lankan artisans with storefronts, product listings, inventory, cart, orders, chat, local payments, reviews, and multilingual support. Artiqo Mobile Marketplace is a Flutter and Firebase mobile app designed for Sri Lankan artisans who need a proper selling platform instead of relying only on social media pages and manual order tracking. The app connects artisan sellers and buyers through dedicated seller profiles, product listings, category browsing, search and filtering, wishlists, cart flows, order management, reviews, ratings, and secure buyer-seller chat. The seller side is not just upload-and-forget: it includes inventory tracking, low-stock alerts, order acceptance/cancellation/completion workflows, and basic sales analytics. The backend shape is Firebase-first: Firebase Authentication handles account access, Firestore stores marketplace data and real-time messages, Firebase Storage supports app assets, and Cloudinary handles product media. The mobile client is built in Flutter and Dart for Android and iOS compatibility. The project is explicitly local-market focused. The README defines Sinhala, Tamil, and English support, local pickup, local delivery support, and Sri Lankan payment gateway integration through PayHere and Genie as part of the product direction. It was developed as Mobile App Development coursework and is still marked as under development in the source repository. ## SplitSmart AI Project URL: https://github.com/deaneeth/splitsmart-ai Skills: AI/ML, React, TypeScript, Google Gemini, OCR, Tailwind CSS Summary: AI-powered bill splitting app using Google Gemini to parse receipts, assign items via natural language, and generate itemized group splits. SplitSmart AI is an intelligent bill-splitting web app that uses Google's Gemini API to turn receipt images into editable, assignable line items. It is built around practical group payment workflows: upload a receipt, add friends, assign items with chat commands or manual controls, then generate a final settlement summary. The app goes beyond basic OCR. Gemini handles receipt extraction, natural-language assignment, dietary classification, and a playful roast mode, while the React interface handles weighted splits, equal split mode, proportional tax/tip distribution, local session history, and shareable export. ## BudgetMe Project URL: https://github.com/deaneeth/budgetme-finance-manager Skills: Full Stack, C#, WinForms, .NET, SQL Server, Desktop App Summary: C# WinForms personal finance manager for tracking expenses, debts, investments, transactions, and financial dashboards. BudgetMe is a C# Windows Forms personal finance manager for tracking expenses, debts, investments, transactions, and financial progress from a local desktop app. It uses .NET Framework 4.7.2, SQL Server LocalDB, and ADO.NET to keep personal finance records on the user's machine. The project is structured as a traditional WinForms solution with separate forms for login, signup, dashboard, daily tracker, debt manager, investment tracking, and transaction management. The README defines the core database tables as users, transactions, debts, investments, and daily expenses, which gives the app a clear local data model rather than just a UI mockup. ## Cosmic Hand Control Project URL: https://github.com/deaneeth/Cosmic-Hand-Control Skills: Full Stack, React, TypeScript, Three.js, React Three Fiber, MediaPipe Hands, Webcam Hand Tracking Summary: Webcam-powered 3D playground where MediaPipe hand gestures select, grab, duplicate, and destroy floating iridescent blocks. Cosmic Hand Control is a browser-based 3D interaction playground where webcam hand gestures control floating iridescent blocks. It combines React 19, TypeScript, Three.js, React Three Fiber, Drei, and MediaPipe Hands to turn a normal webcam feed into a real-time gesture controller. The app tracks 21 MediaPipe hand landmarks, classifies gestures, and maps palm position into the Three.js scene. Open palm selects blocks, pinch grabs and moves them with spring-like physics, fist destroys a selected block with particles, and thumbs up duplicates the selected object. The scene itself uses floating cubes, collision response, stars, fog, lighting, and a mirrored webcam overlay to make the interaction understandable while it runs. ## RepoViz Project URL: https://github.com/deaneeth/RepoViz Skills: Open Source, React, TypeScript, D3.js, Graph Visualization, Google Gemini Summary: AI-assisted repository visualizer that turns a public GitHub repo URL into an interactive D3 graph of files, languages, and dependencies. RepoViz is an AI-assisted developer tool for visualizing a public GitHub repository as an interactive graph. The app accepts a repository URL, asks Gemini 2.5 Flash to return structured JSON, then converts that data into D3 nodes and links representing the repo root, folders, files, languages, and dependencies. The product is built with React 19, TypeScript, Vite, D3, and @google/genai. The useful part is the workflow: AI turns an unknown repo into a typed graph model, D3 renders it with zoom and drag interactions, and the sidebar gives users practical controls for file visibility, dependency filtering, layout mode, node colors, and link styling.