ScriptBreakdown AI — Instant Production Breakdowns

Inspiration

Breaking down a screenplay is one of the most tedious jobs in film pre-production. A 1st AD or line producer manually reads every scene and tags the location, cast, props, time of day, and shoot complexity — hours of work before a single frame is shot. We wanted to hand that grunt work to an AI agent.

What it does

ScriptBreakdown AI takes a screenplay PDF and instantly produces a professional production breakdown: a scene-by-scene table (INT/EXT, location, time of day, characters, props), a cast list, unique locations, a logline guess, and a shoot-complexity note — as a clean, shareable report.

How we built it

A two-agent pipeline powered by Google Gemini:

• Agent 1 — Breakdown: Gemini reads the full screenplay PDF and returns structured JSON (every scene tagged).
• Agent 2 — Report: Gemini turns that JSON into a polished HTML production overview. Built in Python with Flask for the web app, and deployed on the cloud.

Challenges we ran into

Getting strict, parseable JSON out of the model, handling Gemini's high-demand 503s with a retry-and-cache system, and turning a local script into a hosted web app.

What we learned

How to chain multiple LLM agents into a reliable pipeline, prompt for structured output, and ship an agent as a real deployed tool instead of a demo.

What's next

Adding a budget-estimator agent grounded in real-world costs (via the Parallel web-research API), a shooting-schedule/stripboard generator, and export to PDF.

Built With

• ai-agents
• flask
• gemini-api
• generative-ai
• google-cloud
• google-gemini
• llm
• prompt-engineering
• python