Film Production Studio Assistant

Autonomous AI agents parse scripts, query ClickHouse in milliseconds, check inventory, and enforce budgets with Google ADK.

Comment 1

The Story

Every day, film crews lose countless hours to manual administrative bottlenecks. A director wants to shoot a complex scene, but production is stalled. The prop master missed a continuity detail—a specific watch from scene 4. The line producer is frantically recalculating the daily budget on a broken spreadsheet.

"Script breakdowns"—identifying every prop, wardrobe item, and piece of equipment needed for a scene—are traditionally done by hand. While AI has revolutionized video generation, the operational side of filmmaking remains stuck in the past.

What production crews actually need is not another static spreadsheet template. They need an intelligent system that understands the narrative, checks the warehouse, and balances the budget in seconds.

That is the vision behind Agentic Cinema.

What is Agentic Cinema?

The Film Production Studio Assistant is an autonomous multi-agent platform that bridges natural language screenplays with real-time inventory, financial analytics, and production logic.

Rather than relying on manual logging, our system processes script scenes through a pipeline built with the Google Agent Development Kit (ADK) to instantly answer:

  • Do we have these props in stock?

  • Are we exceeding our daily rental budget?

  • Did we forget a continuity item from a previous scene?

Within milliseconds, the platform:

  • Reads the script: Ingests raw screenplay scenes.

  • Extracts entities: Identifies props, wardrobe, and set pieces.

  • Queries inventory: Verifies ClickHouse stock availability and prices in real-time.

  • Enforces rules: A specialized Judge Agent evaluates continuity warnings and enforces strict budget limits.

Who is this for?

Agentic Cinema focuses on helping production teams make better, faster decisions:

  • Line Producers: Automatically generate daily rental budgets without manual data entry.

  • Prop Masters: Instantly cross-reference script requirements with live warehouse inventory.

  • Script Supervisors: Catch continuity errors (e.g., missing holsters or vials) before the cameras even roll.

Platform Architecture

Agentic Cinema follows a serverless multi-agent pipeline orchestrated through Google ADK. .

                        Raw Screenplay (Input)
                                  │
                                  ▼
          ┌───────────────────────────────────────────────┐
          │                                               │
          │             ORCHESTRATOR AGENT                │
          │           (Central Hub & Manager)             │
          │                                               │
          │   3. Uses Skill to query DB with extracted ◄──┼─── ClickHouse
          │      items to get prices and availability     │    (Inventory DB)
          └─────┬──────▲─────────────────────┬──────▲─────┘
                │      │                     │      │
     1. Sends   │      │ 2. Returns          │      │ 5. Returns
     Script     │      │ Items               │      │ Budget/Logic
                │      │ Report              │      │ Report
                ▼      │                     ▼      │
          ┌────────────┴─┐    4. Sends     ┌────────────┴─┐
          │   Agent 01   │    DB Info &    │ Judge Agent  │
          │ (Extractor)  │    Prop List    │ (Evaluator)  │
          └──────────────┘                 └──────────────┘
             (Analyzes text                  (Checks errors
             & extracts props)               & budget limits)
                                  │
                6. Generates Final Consolidated Report
                                  │
                                  ▼
                       Agentic Web Dashboard UI
              (Final Inventory + Judge Recommendations)

How I built it

I built a robust multi-agent pipeline leveraging core cloud and AI technologies:

  • Google Cloud ADK & Gemini Engine: Implemented a modular multi-agent architecture using the official google-adk framework and FastAPI in Python, featuring dedicated agents for extraction and a Judge Agent for business logic.

  • ClickHouse Cloud Analytics: The orchestrator triggers optimized SQL queries against ClickHouse to evaluate inventory levels, item statuses, and financial metrics at lightning speed.

*Replit Hosting: The entire Python backend orchestrator and web interface were developed, configured, and hosted on Replit.

Challenges I ran into

Integrating non-deterministic language outputs from an LLM with strict relational database queries and business rule enforcement was a key challenge. I solved this by structuring a dedicated multi-agent orchestration pattern where the Judge Agent systematically cross-references extracted props against ClickHouse financial limits and continuity requirements. I also resolved module routing and dependency management across the Replit environment.

Accomplishments that I'm proud of

  • Delivering a fully functional, multi-agent application that performs script extraction, database verification, and automated budget/continuity judging in milliseconds.

  • Successfully implementing a specialized Judge Agent via the official google-adk framework to catch critical continuity warnings and enforce budget caps.

  • Integrating ClickHouse Cloud analytics to demonstrate real-time performance across production inventory items.

What I learned

I gained valuable hands-on experience structuring multi-tiered agentic workflows with Google ADK, implementing automated business rule enforcement through AI judges, writing high-speed analytical queries in ClickHouse Cloud, and deploying Python web applications on Replit.

Future Roadmap

Agentic Cinema is designed as a continuously evolving production platform. Future capabilities include:

  • Semantic Protection & Governance: Integrate advanced agent governance and semantic protection to secure the multi-agent architecture against prompt injections and ensure safe corporate-scale operations.

  • External Sourcing Agent: Expand the architecture by adding an autonomous procurement agent. When props are missing from the internal ClickHouse inventory, this agent will search external vendor APIs to generate real-time sourcing options.

  • Multi-Scene Batch Processing: Expand the system to process entire feature-length scripts simultaneously.

  • Automated Shooting Schedule Optimization: Recommend shooting order based on prop availability and daily rental costs.

Impact

Administrative bottlenecks influence millions of everyday decisions on a film set, yet the tools used to solve them remain manual and scattered. Agentic Cinema unifies production intelligence into a single AI-powered platform capable of extracting, reasoning, and recommending. By combining natural language processing with live inventory logic, we are giving filmmakers their most valuable resource back: the time to focus on what actually matters—storytelling

Built With

Share this project:

Updates

Submission history