GreenLit-AI

Inspiration

Every year, film productions lose millions due to historical inaccuracies, continuity errors, and legal issues that could have been caught before filming. We asked: what if AI could research a script the way a team of script supervisors does — but in seconds, not weeks?

We were inspired by the real workflows of production coordinators who manually fact-check scripts scene by scene. Greenlit AI automates that entire process using a multi-agent system where specialized AI agents work in parallel, just like a real production research team.

What It Does

Greenlit AI is a multi-agent orchestration platform that extracts factual claims from film/TV scripts and verifies them in real-time. It deploys 4 specialized AI agents working together:

• Director Agent — Extracts factual claims, historical references, locations, and technical details from scripts
• Research Agent — Verifies each claim with live research using Parallel API
• Legal Agent — Flags copyright, licensing, and clearance risks
• Continuity Agent — Tracks character consistency and timeline accuracy across scenes

Paste a script, and get a full production report with verified claims, flagged inaccuracies, confidence scores, and research sources — all in a cinematic interface built for real production teams.

How We Built It

• Frontend: Next.js 14 with TypeScript and Tailwind CSS, styled with a cinematic "script supervisor's desk" aesthetic — film sprocket borders, marquee lighting effects, and film grain textures
• Backend: Python FastAPI deployed on Google Cloud Run
• AI Orchestration: Google Cloud Gemini Enterprise Agent Platform powers the Director, Legal, and Continuity agents
• Real-time Research: Parallel API handles live fact verification for every extracted claim
• Deployment: Frontend on Vercel, backend on Google Cloud Run

The multi-agent architecture was the key innovation — instead of a single AI doing everything, we built specialized agents that each handle one aspect of production research, then combine their findings into a unified report.

Challenges We Faced

• Multi-Agent Coordination: Getting 4 specialized agents to work in parallel while producing a coherent combined report was the biggest technical challenge. We had to design careful orchestration logic to merge results without conflicts.
• Claim Extraction Accuracy: Training the Director Agent to reliably distinguish factual claims from dialogue, stage directions, and creative fiction required iterative prompt engineering.
• Real-time Research Quality: Filtering noise from Parallel API results and matching research back to specific claims demanded careful parsing and confidence scoring.
• Cinematic Design: Building a UI that feels like a professional production tool — not a generic dashboard — while keeping it fully functional and responsive.

What We Learned

• Multi-agent architectures are powerful but require careful orchestration design
• Real-time API integration (Parallel) adds significant value over static AI responses
• Domain-specific design matters — the cinematic theme makes the tool feel purpose-built for filmmakers
• Google Cloud's agent platform provides a solid foundation for building specialized AI workflows

What's Next

• PDF script upload support for full screenplays
• Batch processing of entire scripts scene-by-scene
• Auto-notifications via Slack/Email for high-risk legal issues
• Differential re-analysis that only processes changed sections on script revisions
• Google Drive integration for automatic script monitoring and processing

Built With

• ai
• fastapi
• google-cloud-gemini
• google-cloud-run
• multi-agent
• next.js
• parallel-api
• python
• tailwind-css
• typescript
• vercel