CineAgent – AI Script & Production Triage Agent

• 

  to do page

• 

  swith between to do (now sected vfx)

• 

  home page with AI chat

• 

  adding document (source)

Inspiration

Film pre-production is notoriously manual and time-consuming. Line producers, assistant directors, and department heads spend days manually reading through 120-page screenplays to break down scenes, extract prop lists, tag stunt requirements, and coordinate wardrobe needs. Built for the Agentic Cinema Hackathon, CineAgent was created to eliminate this manual operational bottleneck by turning raw, unstructured screenplays and production briefs into real-time, automated departmental task boards.

What it does

CineAgent serves as an interactive, internal AI workspace designed specifically for film production crews:

Script & Document Ingestion: Production leads can upload screenplays, call sheets, or location agreements in PDF or TXT format.

Contextual Production Q&A: A Gemini-powered chat agent answers complex logistics and scene breakdown questions directly using context extracted from uploaded documents.

Automated Department Triage: The agent parses raw screenplay text into structured task cards and automatically categorizes them into departmental tabs: Props, VFX, Wardrobe, and Stunts.

Interactive Task Board: Crews can track, update, and manage task statuses across To Do, In Progress, and Done in real time.

How we built it

Platform & Hosting: Scaffolded, developed, and hosted live using Replit with Replit Agent.

Frontend: Built with React and Tailwind CSS, styled in a dark-mode, cinematic theme for high usability on active film sets.

Backend & Database: Powered by a Node.js/Express API with PostgreSQL and Drizzle ORM to manage document records, chat history, and departmental task states.

AI Engine: Integrated with the Google Gemini API to handle deep text parsing and generate structured JSON schemas for seamless departmental task routing.

Challenges we ran into

Prompt Engineering for Multi-Departmental Rules: Fine-tuning Gemini prompts so the model accurately detects implicit departmental needs (e.g., recognizing that "a smoking gun" requires both a Prop entry and a Stunt/SFX safety task) without syntax or formatting errors.

Real-time Deployment & Environment Setup: Managing database schema migrations (via Drizzle ORM) and environment secrets cleanly within the Replit cloud environment while adhering to daily development quotas.

Accomplishments that we're proud of

Successfully building and publishing a full-stack, end-to-end production tool hosted live on a *.replit.app domain.

Creating a sleek, dark-mode user interface that makes complex script breakdown data easily readable for busy production crews.

Seamlessly connecting Gemini's unstructured document parsing directly to real-time PostgreSQL database state updates.

What we learned

How to effectively structure system instructions and JSON schemas for Google Gemini when orchestrating multi-step agentic workflows.

How to rapidly scaffold full-stack architectures, relational schemas, and API endpoints using Replit Agent.

What's next for CineAgent – AI Script & Production Triage Agent

Automated Storyboarding: Integrating text-to-image generation directly into task cards to generate scene visual concepts automatically.

Industry Format Exports: Supporting direct exports to industry-standard pre-production software (such as Final Draft and Movie Magic Scheduling).

Multi-User Collaboration: Adding live WebSocket synchronization so multiple department heads can update shot lists and prop status simultaneously on set.

Built With

• agentic-ai
• ai-agent
• drizzle-orm
• express.js
• film-production
• gemini-api
• google-gemini
• javascript
• node.js
• postgresql
• react
• replit
• tailwindcss
• typescript