Kumuditha Tharinda Liyanage

Desktop application · Solo

Mac cleanup tools are mostly abandonware or upsell funnels; a cleaner has to be honest about what it deletes and fast enough to scan a full disk without blocking the interface.

• Structured the app as MVVM over Swift actors so multi-gigabyte disk scans never freeze the SwiftUI interface.
• Shipped in-app purchases through StoreKit 2 and wired a CircleCI pipeline for continuous integration.
• Managed login items through SMAppService rather than editing launch agents by hand.

• Swift 5.9
• SwiftUI
• Swift Actors
• SMAppService
• StoreKit 2
• CircleCI

github.com/mr-kumuditha/MacPulseCase study

Full-stack web application · Solo

Distributed-system failures are taught as diagrams and experienced as outages; there is no safe place to rehearse a request that succeeded but whose reply was lost.

• Built sandbox inventory, payment and delivery services so faults happen over real HTTP rather than mocked calls.
• Implemented idempotent recovery — keys, retry with backoff, deduplication — with a baseline-versus-recovery comparison view.
• Exported every run as JSON evidence and covered the fault-injection flows with Playwright.

• TypeScript
• Next.js
• Node.js
• SQLite
• Playwright
• Docker

rehearsal.tharinda.devgithub.com/mr-kumuditha/rehearsalCase study

HND final project · Cross-platform mobile · Group project

Global trip planners know nothing about local roads or coverage, leaving travellers with a list of names, no order and no map once they lose signal.

• Built itinerary generation on Gemini through Firebase AI Logic with App Check, removing client-side model keys.
• Migrated the integration off a Cloud Functions gateway to Firebase AI Logic directly and verified it on a physical device.
• Routed between stops with OSRM and cached map tiles so itineraries stay readable offline.

• Flutter
• Dart
• Firebase AI Logic
• Gemini
• OSRM
• Supabase

github.com/mr-kumuditha/NavoraCase study

Campus project · Native Android · Solo

A repair shop runs on four people wanting four views of the same job, and building four apps was not an option on a two-day assignment deadline.

• Modelled one data layer with role-scoped views for customers, technicians, branch managers and administrators.
• Matched customers to the nearest branch by GPS and tracked repair status live on Firebase.
• Delivered a production-style workflow — design tokens, on-device testing, structured git history — inside the deadline.

• Kotlin
• Jetpack Compose
• Firebase
• Supabase
• SQLite

github.com/mr-kumuditha/fixora-mobile-appCase study

Campus project · IoT · Team of 3

Underground, a supervisor cannot see where a worker is or what they are breathing until it is already an incident report.

• Wrote ESP32 firmware in C++ polling MQ-135 gas, DHT11 temperature and MPU6050 motion, with MFRC522 RFID identifying the wearer.
• Streamed telemetry to Firebase Realtime Database and fired Telegram alerts on threshold breaches.
• Built the TypeScript supervisor dashboard showing live zone, gas and status across active helmets.

• ESP32
• C++
• Firebase Realtime Database
• TypeScript
• Telegram Bot API

minelivemonitor.systemsgithub.com/mr-kumuditha/IoT-Safety-Monitoring-SystemCase study

Test IP condition generation at unsigned address boundaries. Added direct coverage for IPCondition.getCondition() in the API module. Existing gateway and DAO tests exercise conditions indirectly; this checks the IPv6 Siddhi output at unsigned address boundaries.

• Java
• Siddhi
• JUnit

Pull request #13986

Explain Retry-After handling for throttled requests. The rate-limit error guide lists error codes but does not explain how consumers should read the Gateway's Retry-After header — a client assuming an integer delay cannot parse the HTTP-date value it can emit.

• Markdown
• API Manager

Pull request #11857Pull request #11858