Mobile Software Architect @ Vodafone · Android · AI-Assisted Development
Istanbul, Türkiye
I design and evolve large, multi-module Android codebases, and build the tooling that lets engineering teams use AI safely in their day-to-day work: reviewing code, keeping architecture consistent, and shipping faster without lowering the quality bar.
Modular by design. AI-assisted by default.
I work on the architecture of Yanımda, Vodafone Türkiye's self-service Android app, a large multi-module Kotlin codebase.
- AI pull-request reviewer (IntelliJ / Android Studio plugin). Reviews GitHub Enterprise PRs with a multi-model consensus. It produces a checklist, a summary, inline comments and architectural issues, and handles batching, streaming, retries and truncated-response recovery for large PRs.
- AI-ready repository. Built a repo-level skills pipeline (
AGENTS.md, scoped instructions, delivery skills) and an[AI]commit convention, so AI-assisted changes are consistent and traceable. - UI modernization. Driving the XML → Jetpack Compose migration with reducer-based MVI as the standard for new screens, with performance benchmarks as guardrails.
- Conversational AI POC. Delivered a proof of concept for proactive, in-app assistant (TOBi) interventions; it was approved to move forward.
- Architecture & mentoring. Enforce module boundaries (
app → feature → domain ← data), write ADRs and review guidelines, and mentor developers across the team.
| Project | What it does | Stack |
|---|---|---|
| Pull_Request_Analyzer (private) | Web app that reviews GitHub / GitHub Enterprise PRs with Claude or DeepSeek and posts line-level review comments | React · TypeScript · Node |
| Skills-Architecture (private) | A portable collection of 31 Claude skills for native Android, iOS and KMP development: architecture decisions, templates, quality gates and a delivery pipeline | Claude Code · Markdown · Shell |
| MobiQA (private, in progress) | Autonomous mobile QA: AI-driven exploratory testing plus accessibility checks and statistical performance-regression detection | Python · Kotlin · Firebase Test Lab · Perfetto |
| PR Memory Engine (design) | Turns the why behind merged PRs into curated, human-approved project memory that lives in the repo | LLM agents · Git |
| Code-health dashboard (private) | Tracks code smells, modularization, Compose adoption and test coverage of an Android codebase over time | Flask · React · SonarQube |
Every non-trivial change an AI agent makes runs the same gated route: clarify, route to the right skills, get the plan approved, build in phases, test, document, verify, then commit and open a PR. Hard gates stop the flow until they are met, and optional checks (UI tests, benchmarks, security) run only when the change calls for them.
- Block Garden: Zen Puzzle (private): an 8×8 block puzzle game with a sticker-collection meta layer, built with Unity 6 under Zen Leaf Games. Uses Firebase Crashlytics, Remote Config, Cloud Save and IAP.
- My Hubs (private): a native macOS app (Swift + WKWebView, stdlib Python server) that starts, stops and tails logs for all my local dev projects from a single window.
- Local generative AI: an on-device image generation setup with ComfyUI, FLUX.2 Klein and Ollama (Qwen) on Apple Silicon, for prompt engineering and photo-editing workflows.
- ComposeCodeBase (private): a multi-module Jetpack Compose reference app with Gradle convention plugins.
Mobile
AI & tooling
Web & scripting
Open to conversations about Android architecture, AI-assisted engineering workflows and developer tooling. Reach me on LinkedIn.