This repository contains Stratum, a free and open-source two-factor authentication (2FA) app for Android with Wear OS support. It implements TOTP, HOTP, Steam, Yandex, and mOTP code generation algorithms with configurable parameters. The app features encrypted local backups with AES-GCM/CBC and Argon2id/PBKDF2 key derivation, import from over 15 third-party authenticator apps (Aegis, Bitwarden, KeePass, etc.), QR code scanning via CameraX, and a Material Design 3 UI. The architecture follows a clean layered pattern with a core library (Stratum.Core) containing domain entities, OTP generators, backup converters, and service interfaces, a Droid project for the phone UI, and a Wear OS project. The codebase is well-organized with dependency injection (Autofac), repository pattern for SQLite persistence, and unit tests. It demonstrates strong data engineering skills in reverse-engineering and parsing various backup formats, implementing custom cryptography for Ente Auth, and maintaining a large icon database. The project is actively maintained, targets .NET 10, and has localization in 27+ languages.
Androidify is a modular Android application that allows users to create personalized Android robot avatars using AI. It accepts a photo (from camera or gallery) or a text prompt, then generates a custom Android mascot image via a fine-tuned Imagen model through Firebase Vertex AI, with on-device Gemini Nano as a fallback for prompt generation and image validation. The app uses ML Kit for pose detection in the camera flow and subject segmentation for background removal. It features adaptive Jetpack Compose UIs across phones, tablets, foldables, and Android XR with shared element transitions. A companion Wear OS app enables one-tap installation of the generated avatar as a watch face, including APK creation and transfer via the Wearable Data Layer. The architecture is multi-module with custom Gradle plugins, Hilt DI, and feature flags via Firebase Remote Config. It is primarily an applied AI sample app, not a data research project, though it heavily relies on configurable AI prompts and model names.