Netflix GPT

Problem

The base curriculum project (Namaste React by Akshay Saini) stops at browsing a movie catalog behind Firebase auth — a solid clone, but not something that demonstrates shipping a real feature end-to-end or getting a production build actually deployed cleanly.

Approach

Extended it in two directions: added a genuine AI feature on top of the clone — a GPT-style search where you describe a mood or a vague prompt and Gemini returns real movie recommendations — and treated deployment as part of the build, not an afterthought, wiring up GitHub Actions CI/CD (TypeScript check + build on every PR) rather than shipping straight from a local machine.

Key decisions

• Used Redux Toolkit for global state (auth, movie data, GPT results) instead of scattering fetches across components, since the GPT search results needed to be readable from multiple screens.
• Integrated Gemini 2.5 Flash Lite specifically for the mood-based search, rather than keeping the project as a plain catalog browser — this is the one feature that isn't in the base curriculum project.
• Hit and fixed a real CRLF line-ending issue breaking the Linux-based Vercel build (the project was developed on Windows) — normalized it via .gitattributes instead of a one-off workaround.

Outcome

// TODO: Coursework-derived project with custom extensions, not a production app with real traffic — no usage metrics to report.