Android Offline AI Note-Taker
Budget: ₹600 – ₹1,500 INR
I’m building a truly offline-first note-taking experience for Android. The app will receive audio snippets (WAV only) sent from a wearable pendant, save them locally, then handle everything on-device:
• Transcribe the recording as soon as it arrives.
• Pass the text through a compact on-device LLM to create a short, coherent summary.
• Store both the transcript and the summary in a local vector index so I can run keyword searches or open a Chat-style interface to ask questions about my past notes.
No cloud calls are allowed at any point—airplane mode must not break any function. I’m comfortable with solutions such as Whisper.cpp or Vosk for speech-to-text and llama.cpp or similar for the LLM, as long as the final APK keeps all models on the handset and latency stays reasonable (≈5 s to process a 30 s clip on a mid-range Snapdragon).
Deliverables
1. Full Android Studio project with build instructions.
2. Pre-trained models and a script or README explaining how to swap/upgrade them.
3. Signed release APK for quick testing.
4. Brief performance report: average processing time, disk footprint, and battery impact for a 30 s WAV file.
Acceptance criteria
• Works completely offline.
• Correctly processes WAV input, producing accurate transcripts and logical summaries.
• Search and chat return relevant results in under 1 s on-device.
• Code is clean, documented, and compile-ready.
If you have proven experience embedding Whisper, llama.cpp, or similar tech into Android, I’d love to see a quick prototype or demo reference along with your bid.
• Transcribe the recording as soon as it arrives.
• Pass the text through a compact on-device LLM to create a short, coherent summary.
• Store both the transcript and the summary in a local vector index so I can run keyword searches or open a Chat-style interface to ask questions about my past notes.
No cloud calls are allowed at any point—airplane mode must not break any function. I’m comfortable with solutions such as Whisper.cpp or Vosk for speech-to-text and llama.cpp or similar for the LLM, as long as the final APK keeps all models on the handset and latency stays reasonable (≈5 s to process a 30 s clip on a mid-range Snapdragon).
Deliverables
1. Full Android Studio project with build instructions.
2. Pre-trained models and a script or README explaining how to swap/upgrade them.
3. Signed release APK for quick testing.
4. Brief performance report: average processing time, disk footprint, and battery impact for a 30 s WAV file.
Acceptance criteria
• Works completely offline.
• Correctly processes WAV input, producing accurate transcripts and logical summaries.
• Search and chat return relevant results in under 1 s on-device.
• Code is clean, documented, and compile-ready.
If you have proven experience embedding Whisper, llama.cpp, or similar tech into Android, I’d love to see a quick prototype or demo reference along with your bid.