Solo dev builds AI‑powered movie tracker for households

A full‑stack developer based in the Netherlands spent the past year turning evening coding sessions into a public Android application called I Like Movies. The app launched on Google Play in the summer and is designed for multiple people living together to maintain a single library of movies and TV shows, track what has been watched, and receive recommendations through a conversational interface.

The technical foundation is deliberately conservative. The backend runs as a single Go binary, exposing a GraphQL API built with gqlgen, while data lives in a PostgreSQL database that also handles full‑text search and import staging. The mobile client is written in React Native using Expo, which allows over‑the‑air updates to be pushed instantly without waiting for Play Store review. Film and television metadata are sourced from TMDB, keeping the stack simple and maintainable for a solo maintainer.

The motivation behind the product stems from a common frustration: most movie‑tracking services assume a single user, making it difficult for couples or roommates to coordinate viewing choices. By sharing one watchlist and viewing history across devices, the app eliminates the need for each person to remember what the other has added, turning the decision‑making process into the primary value proposition rather than search capabilities.

Choosing GraphQL over a traditional REST architecture was a pivotal decision. The app presents many different screens, film details, friend profiles, grouped watchlists, and statistics, each requiring a distinct shape of data. With GraphQL a single schema can satisfy all these needs, avoiding the proliferation of endpoints that a REST design would entail. The generated type definitions also enforce consistency between the mobile and server codebases, reducing runtime errors.

Running an LLM feature as a one‑person operation introduced specific operational concerns. Guardrails such as rate limits, per‑user caps, and input validation were implemented before the public launch to prevent abuse of the chat endpoint tied to the developer's API key. Maintaining the service alone also means that scaling beyond a modest user base could become costly, and any major outage would fall entirely on the creator to resolve.

The AI component works by delegating all factual work to deterministic code while the language model handles natural‑language understanding. When a user asks for a recommendation, the model generates a query that is resolved against TMDB, cross‑checked with the user's watch history to avoid repeats, and filtered for streaming availability in the user's country. An additional quirky feature lets users snap a photo of a DVD shelf or a screenshot of a recommendation list; the system extracts titles from the image and offers to add them to the shared collection after user confirmation.

Future plans include an iOS version currently under development and an open invitation for user feedback to guide further improvements. The developer emphasizes that the app remains free, with the belief that real‑world usage and community input will drive its evolution faster than isolated roadmap speculation.