Technical Lead - Android with 18+ years building production software. I work at the intersection of Android engineering (14+ years) and applied machine learning, with a focus on running models on the device rather than behind an API call.
Currently interested in: on-device inference, agentic systems, and the hard problem of testing AI-generated and state-heavy UI code.
Production Android Kotlin, Jetpack Compose, Coroutines, multi-module architecture. I lead a small Android team and own the technical direction for our mobile platform, including the AI feature track that shipped to production.
On-device ML TensorFlow Lite / LiteRT, quantised models, streaming audio pipelines, VAD and TTS on constrained hardware. I've shipped features that work with no network at all, which is a different engineering problem from calling a hosted model.
Agentic AI & testing Most of my open-source time goes here. Agents that write code are easy; knowing whether the code is correct is not. That gap is what I build tools for.
I build tools to answer research questions, then write up what they find.
| Project | What it does |
|---|---|
| AgentTest | Gradle plugin that generates metamorphic state-restoration oracles for Jetpack Compose. Core finding: remember vs rememberSaveable defects are omission faults, structurally invisible to mutation testing. |
| HotFixAgent | Agentic pipeline for automated Android defect repair. Paper under review at ICTer 2026 (Springer CCIS). |
| ProdCheck | Detekt-based code review CLI with an LLM layer. Caught an LLM incorrectly justifying a requireNotNull substitution as behaviour-identical to !!. |
A paper on metamorphic state-restoration oracles for Android is in preparation for FORGE 2027.
Apps on Google Play, built and maintained independently:
- Ceylon Calendar — 100,000+ downloads
- AtharaMaga — driving companion with geofenced proximity alerts, Sinhala TTS, and Android Auto integration
- NutriAyu — nutrition and Ayurvedic guidance
- ExpensesBook — personal expense tracking more -> https://play.google.com/store/apps/developer?id=Ceylonapz+Technologies
Most of my day-to-day engineering lives in private repositories at work and under client NDA, so the contribution graph here is not a good proxy for what I actually build. The research tooling above is the part I can talk about openly, and it's the part I'd want to be judged on anyway.
- MSc in Information Technology (Enterprise Applications Development), SLIIT
- BSc (Hons) Computing, University of Greenwich
- Founder, CeylonApz
Open to conversations about on-device ML, agent reliability, and anything involving Kotlin and a deadline.