Quantum Recommender

Inspiration

We originally wanted to build a hybrid recommender that could handle movies, music, and books. With only a week to work, we narrowed the scope to movies. Recommendation systems map naturally to QUBO optimization, so this was a perfect chance to try QAOA on a real application.

What it does

Quantum Recommender takes a user’s movie ratings and generates personalized recommendations using a hybrid classical and quantum pipeline. It predicts classical relevance, builds a diversity-aware QUBO, solves it with QAOA, and returns a curated set of movies in a simple frontend.

How we built it

For the frontend, Eli created the React interface for rating movies and viewing recommendations, implemented classical scoring using cosine similarity, cleaned and formatted the dataset, and integrated TMDB poster lookups.

For the backend, Krish formulated the recommendation problem as a QUBO, implemented QAOA using Qiskit and SPSA, built the hybrid solver that combines quantum output with classical scores, and developed the Flask API that links everything together.

Challenges we ran into

We struggled with tuning QAOA so it converged consistently, dealt with normalization bugs that caused classical ratings to collapse, ran into CORS issues and mismatched JSON responses, spent time cleaning inconsistent movie titles, and had several environment problems such as missing Qiskit packages.

Accomplishments we’re proud of

We built a fully working hybrid quantum–classical recommender system, successfully applied QAOA to a real optimization task, created smooth communication between the React frontend, Flask backend, and Qiskit pipeline, and solved a large number of integration issues under a tight time limit.

What we learned

We learned how to build practical QUBO and Ising models, how to tune QAOA with SPSA, how classical similarity metrics behave, how to integrate frontend, backend, and quantum workflows, and how to debug complex systems efficiently.

What’s next for Quantum Recommender

Next steps include expanding to a larger movie dataset, running QAOA on real quantum hardware, adding support for additional media types such as music and books, improving the hybrid scoring strategies, and enhancing the UI with more personalization features.