AI Music Snippet Recommender Prototype
Budget: ₹100 – ₹400 INR
I’m building an Android-only music app where a listener can highlight any six-second to thirty-second slice of a song and instantly receive AI-driven suggestions that sound alike. The flow is deliberately simple: scrub the waveform, tap “Compare”, and a carousel of matches appears. Behind that single tap I want a content-based model that extracts timbre, tempo, key and mood vectors from the selected section, searches a local cache or cloud catalogue, then streams short previews for fast sampling.
Here’s what I need from you:
• A clickable Android prototype (Kotlin or Flutter are both fine) showing the waveform selector, “Compare” button, and a results view. No account system or playback rights management yet—just mocked audio files are enough for this stage.
• A lightweight REST API stub that accepts an audio clip or timestamp reference, calls an AI micro-service, and returns ranked track IDs plus confidence scores.
• A proof-of-concept AI service: Python notebook or small container that uses librosa or similar DSP tools to extract features, compares them with cosine similarity, and feeds the API. I’m happy with pre-computed embeddings for speed.
• Simple SQLite or Firebase setup to store track metadata and the pre-computed feature vectors.
• Clean, material-style UI: dark theme, large touch targets, nothing more than play/pause, scrubber, and results carousel.
Acceptance criteria
1. Select any segment of a test track and receive at least three recommendations within three seconds on a mid-range Android phone.
2. Similarity scores and basic metadata (title, artist, album art) render correctly.
3. Code is push-button deployable and documented enough for me to extend the model later.
If you’ve worked with audio feature extraction, TensorFlow Lite, or on-device ML kits before that’s a plus, but clarity and maintainability matter most. Let me know how you’d tackle the segment analysis and what open-source libraries you’d leverage so we can keep iteration fast.
Here’s what I need from you:
• A clickable Android prototype (Kotlin or Flutter are both fine) showing the waveform selector, “Compare” button, and a results view. No account system or playback rights management yet—just mocked audio files are enough for this stage.
• A lightweight REST API stub that accepts an audio clip or timestamp reference, calls an AI micro-service, and returns ranked track IDs plus confidence scores.
• A proof-of-concept AI service: Python notebook or small container that uses librosa or similar DSP tools to extract features, compares them with cosine similarity, and feeds the API. I’m happy with pre-computed embeddings for speed.
• Simple SQLite or Firebase setup to store track metadata and the pre-computed feature vectors.
• Clean, material-style UI: dark theme, large touch targets, nothing more than play/pause, scrubber, and results carousel.
Acceptance criteria
1. Select any segment of a test track and receive at least three recommendations within three seconds on a mid-range Android phone.
2. Similarity scores and basic metadata (title, artist, album art) render correctly.
3. Code is push-button deployable and documented enough for me to extend the model later.
If you’ve worked with audio feature extraction, TensorFlow Lite, or on-device ML kits before that’s a plus, but clarity and maintainability matter most. Let me know how you’d tackle the segment analysis and what open-source libraries you’d leverage so we can keep iteration fast.