A PWA for finding films and series that uses user taste to evolve recommendations over time.
The problem
Sglitch addresses a simple but real issue: choosing what to watch is difficult not because content is scarce, but because there is too much choice and recommendations are too generic.
Traditional engines mainly use genre, popularity, cast or similarity with titles already watched. Sglitch progressively builds a richer representation of taste: what users save, discard or consider interesting, and the patterns emerging across films and series.
The solution
The solution is an AI-based film discovery engine combining:
- TMDB API for structured film and series data;
- Claude API for interpreting preferences, relationships and softer signals;
- a user profile that evolves through interactions;
- lists and interest signals that feed future recommendations;
- a UI for exploring the catalogue beyond title or genre search.
What I built
I built a React and Vite experience with Node and Express APIs, MySQL persistence, and Claude and TMDB integrations.
How it works
Discovery > interest signals > taste profile > ranking and recommendations > personal lists > persistent memory and avatar-level evolution.
Engineering choices
- React and Vite support an interactive interface with profiles, lists, interest signals and dynamic navigation.
- Node.js and Express orchestrate frontend requests, database access and external APIs without exposing credentials.
- MySQL preserves profiles, lists, signals and interactions so the system can evolve over time.
- TMDB API supplies structured catalogue data; Claude API interprets relationships between preferences and titles.
- Data and interpretation stay separate: TMDB describes what exists; Claude helps explain why it may interest a person.
- Persistent, evolving profiles turn saved items, rejections and lists into future recommendations.
- PWA makes the product feel installable and persistent; PostHog monitors activity in the app.
Outcome
A discovery product that uses every interaction to improve the user profile, not merely to return search results.
