MOVIE-INATOR

Inspiration We were inspired by the movie Odyssey, after watching it, we got curious about everything that actually goes into producing a film. Reading up on how producers break down a script (into scenes, characters, locations, props, and every other production element) showed us how much fragmented, easy-to-lose information a production team has to reconcile before a single decision gets made.

What it does Movie Inator is a producer workspace: you upload a script plus the team's supporting notes (location access, schedule and budget assumptions, department notes, etc.), and it turns them into one read-only Producer Intake Decision Packet a checklist showing what the film needs, what's still unconfirmed, where different documents disagree, and who needs to do what next. Every claim it makes is tied to an exact citation in the source material, so nothing gets stated as fact unless a document actually says it.

How we built it The backend is a Node.js pipeline that ingests PDF/text sources, extracts and classifies content (explicit fact, inference, assumption, conflict, or open question), and assembles it into a versioned, schema-validated packet with a deterministic, credential-free "Demo mode" as the default and an optional Gemini-backed path gated behind explicit server-side configuration. A lightweight browser frontend lets producers upload sources, label them, review the packet by section, and export a read-only handoff.

Challenges we ran into We initially wanted to also match real actors to script characters and generate a budget breakdown, but budgets vary too much by country to do responsibly without picking a location, so we scoped that out. More broadly, we underestimated just how much a script breakdown actually touches, characters, scenes, locations, props, costumes, vehicles, animals, sound, music, stunts, VFX, weather, time of day and had to be disciplined about never letting the system state something as confirmed (a booked location, a contracted actor, an approved stunt) when the source material only implied it.

Accomplishments that we're proud of Our final project is able to identify characters correctly, but of course it still needs more improvement.

What we learned How to work as a team, how to use AI to build a project (vibe coding), and how to collaborate on a large-scale project using GitHub.