Inspiration
The idea came from personal frustration. We’d often spend 20–30 minutes scrolling through Netflix or Spotify only to end up rewatching or replaying the same old favorites. While platforms have their own recommendation engines, they rarely cross over. We thought — what if a single app could understand your personality, taste, and current mood, and offer suggestions across all types of media?
What it does
Recomendo is an AI-powered recommendation platform that provides personalized suggestions for books, movies, and music — all in one intuitive app.
It learns your preferences over time using:
- Your input (mood, genre, interests)
- Previous likes/dislikes
- Conversational context
How we built it
We built Recomendo with a user-first mindset, prioritizing speed, relevance, and personalization:
- Frontend: React (with TailwindCSS for quick UI prototyping)
- Backend: Node.js + Express with MongoDB to store user preferences and feedback
- AI Integration: OpenAI GPT-4o powers the conversational recommendation engine
- APIs Used:
- 🎬 TMDB API for movie data
- 🎵 Spotify API for music genres and tracks
- 📚 Google Books API for book recommendations
We also designed a feedback system, allowing users to like/dislike recommendations, improving results over time.
Challenges we ran into
- API Limitations: Different APIs have different rate limits and data formats, making unification a challenge.
- Mood Detection: It was tough to translate a user's "mood" into relevant recommendations consistently — we had to rely on NLP sentiment cues.
- Time Management: With just one month and many features to pack in, prioritizing the core experience was key.
- Personalization Logic: Balancing AI recommendations with stored user preferences (without overfitting) took experimentation.
Accomplishments that we're proud of
- Unified Multi-Media Recommendations: Successfully integrated recommendations for movies, books, and music in a single platform — something even major apps rarely offer together.
- Conversational AI Integration: Built a smooth, natural chat interface using GPT-4o to handle user queries and return relevant suggestions in real-time.
- Smart Feedback Loop: Implemented a dynamic feedback mechanism that tailors future recommendations based on user likes/dislikes.
- Cross-API Integration: Merged data from TMDB, Spotify, and Google Books into a cohesive system without overwhelming the user.
- Polished MVP in Time: Delivered a working prototype within the tight hackathon deadline — fast, functional, and scalable.
What we learned
- Prompt Engineering Matters: Small changes in how we asked GPT for recommendations made a big difference in quality.
- API Rate Limits and Reliability: Real-world APIs have usage caps and response inconsistencies — caching and fallback logic were essential.
- Personalization Logic is Tricky: Balancing short-term preferences with long-term user profiles required experimentation and data tuning.
- UI/UX Focus is Key: Even a smart engine needs to be wrapped in an intuitive, fast interface for users to stay engaged.
- Rapid Prototyping Techniques: Learned to prioritize core features under time pressure, and iterate fast using test users.
What's next for Recomendo
- User Profiles & Social Sharing: Let users create profiles, save favorites, and share recommendations with friends.
- Mood-Based Discovery: Add real-time mood detection from text or even voice tone to generate smarter suggestions.
- Daily Smart Feed: A personalized daily digest combining one book, movie, and song tailored to your interests.
- Browser Extension & Mobile App: Bring Recomendo to where users are — Chrome, iOS, and Android.
- Collaborative Filtering: Use data from similar users to offer even deeper, more accurate recommendations.
- Gamification: Reward users with badges for trying new genres or completing reading/watching/listening streaks.
Log in or sign up for Devpost to join the conversation.