NeuroV — Offline AI Assistant & Automation (Android)

Siddhesh Sonar
Published

0
  • Project creator

What it is ?
NeuroV is a privacy-first Android assistant that runs fully on-device. It listens, understands, and executes actions without sending data to servers.

What I built ?
Dynamic plugin architecture (Kotlin + Jetpack Compose) to add features without app updates — plugins render their own UI.
Speech pipeline: Whisper STT → Kotlin intent parsing → JSON command execution via Accessibility API.
RoomDB for context/history, WorkManager for resilient background tasks.
C++/JNI bridges for on-device inference (TFLite/ONNX / llama.cpp integration).
CI/CD (GitHub Actions): signed AABs, versioning, release notes.

Highlights
Plugin load time: <250 ms → instant feature enablement.
Zero cloud calls → maximum privacy & offline reliability.
Robust error handling and rollback for “transaction-like” flows.

Tech
Kotlin, Jetpack Compose, Coroutines/Flow, Hilt, Room, Retrofit/OkHttp, WorkManager, AccessibilityService, TFLite/ONNX, JNI/C++, GitHub Actions.