FilmGuru: Your AI-Powered Movie Companion 🎬

FilmGuru’s journey begins with my movie database, a CSV file sourced from Kaggle, containing 8,000 movie records with details like descriptions, year, and other metadata. Here’s how it all comes together:

I started by uploading the CSV to Timescale as a new service, which took just 2 minutes to complete.

Using the SQL editor, I added a primary key column to the table—this is crucial for utilizing Timescale’s pgvector functionality.

I then used pgai with OpenAI’s embedding model (small) to generate embeddings instantly across the entire database, creating a vector representation for each movie entry to allow FilmGuru to find the best matches based on your preferences.

Behind the Scenes of Your Recommendation Journey

When you enter a prompt (like "I want something like Inception"), it’s sent to the backend server.

The server uses pg AI to create a unique embedding for your query with the help of OpenAI's text-embedding-3-small, transforming it into a vector.

FilmGuru then uses pgvector to search the database for movies with similar embeddings, delivering the closest match as your next recommendation.

It’s seamless, fast, and incredibly fun!

Final Thoughts

The project came up smooth. I only had one dependency in my back end. That was the pg library for postgres. Kinda cool that I could build the entire backend just with one library.

I’m planning to add an agent that fetches images for each movie, making recommendations even more visually engaging. Stay tuned!

All the Extensions