Moodify AI 🎵

Feel it. Find it. Play it.

Moodify AI reads how you're feeling from a few sentences of plain text and recommends a playlist to match — powered by a custom-trained emotion classifier and Last.fm's music catalog.

🔗 Live demo: moodifyai-feel-it-find-it.vercel.app

---

How it works

1. You type how you're feeling into the Detector.
2. The text is embedded using sentence-transformers (all-MiniLM-L6-v2) and classified into one of six moods by a LogisticRegression model trained on the GoEmotions dataset.
3. The detected mood is mapped to genre tags, which are sent to the Last.fm API to pull matching tracks.
4. Duplicate tracks are filtered out, the list is shuffled, and the top 10 songs are returned and rendered as a playlist.

Moods detected: Happy, Sad, Angry, Calm, Romantic, Anxious

---

Tech stack

Frontend

• React (Vite)
• Custom scroll-driven scene architecture — a shared ScrollContext + scroll-progress hook drives section transitions with plain JS math (no GSAP/Lenis/animation libraries)
• Vanilla CSS

Backend

• FastAPI
• sentence-transformers for text embeddings
• scikit-learn (LogisticRegression) for mood classification, persisted with joblib
• Last.fm API for track recommendations
• Deployed on Hugging Face Spaces (Docker SDK)

---

Project structure

moodify-ai/
├── frontend/
│   ├── src/
│   │   ├── App.jsx              # Scene layout + nav dots
│   │   ├── ScrollContext.jsx    # Shared scroll progress provider
│   │   ├── useScrollProgress.js # clamp / remap / lerp helpers
│   │   ├── components/
│   │   │   ├── Navbar.jsx
│   │   │   ├── Hero.jsx
│   │   │   ├── MoodsSection.jsx # Auto-playing mood carousel
│   │   │   ├── Detector.jsx     # Text input → mood + playlist
│   │   │   └── AboutFooter.jsx
│   │   ├── assets/images/       # Mood GIFs
│   │   ├── index.css
│   │   └── main.jsx
│   └── package.json
│
└── backend/
    ├── main.py                  # FastAPI app (/detect-mood, /recommendations)
    ├── emotion_song_model.pkl   # Trained classifier
    ├── calm_sentences.txt       # Extra training data (class imbalance fix)
    ├── requirements.txt
    ├── Dockerfile
    └── README.md                # Hugging Face Spaces config

---

Running locally

Backend

cd backend
pip install -r requirements.txt

Create a .env file:

LASTFM_API_KEY=your_lastfm_api_key

Run the server:

uvicorn main:app --reload --port 7860

Frontend

cd frontend
npm install
npm run dev

Update the backend URL in Detector.jsx if you're not using the deployed Hugging Face endpoint.

---

API

POST /recommendations

Request

{ "text": "I can't stop smiling today" }

Response

{
  "mood": "happy",
  "songs": [
    { "name": "...", "artist": "...", "url": "..." }
  ]
}

POST /detect-mood

Returns just the predicted mood for a given text, without song recommendations.

---

Deployment

• Backend — Hugging Face Spaces (Docker SDK), exposed on port 7860. See backend/Dockerfile.
• Frontend — Vercel.

---

Built by Samidha

I believe music should meet you where you are, not where an algorithm thinks you should be.