Production Passport

AI agent that turns film production descriptions into complete reports in seconds. Integrates real-time search, Gemini, geocoding for locations, permits, costs, and film commission contacts.

Comment

Inspiration

As a filmmaker and AI consultant who has worked with 500+ businesses across Mexico, US, and Costa Rica, I kept seeing the same problem: every international film shoot requires days of manual research — permit requirements, film commission contacts, drone regulations, crew costs, insurance. Information scattered across dozens of government websites, often outdated.

When this hackathon launched with Parallel Web Systems + Google Gemini as partners, I knew exactly what to build: an AI agent that automates the entire research workflow. Instead of filmmakers spending hours searching government websites, they just describe their shoot and get a complete production report in seconds.

What it does

Production Passport is an AI agent that takes a simple description of a film production and generates a complete production report in seconds.

Input example:

"We need to shoot in Los Angeles with 10 crew, 5 actors, 20 extras, drones, budget $50,000"

Output includes:

  • 📍 Automatic location detection from coordinates or text (reverse geocoding via BigDataCloud)
  • 📋 Permit requirements with current costs and processing times
  • 💰 Bring vs Hire cost comparison with actual crew/equipment day rates
  • 🏢 Film commission contacts (email, phone, website) — real data via Parallel Search
  • 📱 Drone regulations and restrictions
  • 🛡️ Insurance recommendations
  • ✅ Pre-production checklist
  • 📄 Exportable to DOCX format

Key differentiator: Multi-location comparison — compare California vs New York, Mexico vs Colombia, etc. with data specific to each location.

Live mode: When APIs are configured, Parallel searches the web in real-time for current permit costs, regulations, and vendor contacts. Gemini generates the report from this live data. Without APIs, it works in demo mode with pre-loaded reference data.

How we built

Tech Stack

  • Backend: Python + Flask (REST API on port 8080)
  • Search: Parallel AI Search API (real-time web research) — integrated at runtime
  • AI: Google Gemini API via google-generativeai SDK (Google Cloud SDK requirement)
  • Geocoding: BigDataCloud reverse geocoding API (free, no key required)
  • Export: python-docx for Word document generation
  • Frontend: Vanilla HTML/CSS/JS with Tailwind CSS

Architecture

  1. User submits production description via web UI
  2. Backend extracts production details using regex (crew size, extras, budget, drones)
  3. Reverse geocoding detects exact location from coordinates
  4. Parallel Search API queries for: permit costs, crew rates, film commission contacts, regulations
  5. Gemini generates a structured, natural-language report from the collected data
  6. Report displays in the browser with formatted sections
  7. User can download as DOCX

Real-time data flow (Live mode)

  • Parallel searches:
    • "[Location] film permit costs 2025"
    • "[Location] film crew day rates"
    • "[Location] film commission contact email"
  • Results are parsed for prices, contacts, and links
  • Gemini receives all this data and generates a comprehensive report

Demo mode

Without API keys configured, the system uses pre-loaded reference data for major production hubs (California, New York, Mexico, Colombia, Spain, Japan, UK) — ensuring the demo always works.

Challenges

  1. Gemini API model versions: The model names kept changing (gemini-1.5-flash → gemini-2.5-flash → gemini-3.6-flash). Had to test multiple versions and update the code to use the current available model.

  2. Gemini timeout issues: Gemini API was hanging on requests. Fixed by reducing timeout to 15 seconds and adding proper User-Agent headers. Added graceful fallback to demo mode when API fails.

  3. Parallel API key configuration: Had to verify the API key was valid and had proper permissions. Sandbox vs production keys behave differently.

  4. Multiple location detection: Initial code only detected the first state in "California vs New York" — had to fix to extract and search for ALL mentioned locations.

  5. Comparison consistency: Gemini wasn't consistently including comparison data across all report sections. Fixed by restructuring the prompt to explicitly require per-location data in each section.

  6. Google Cloud billing is complex: Understanding the spending limits, credit requirements, and tier upgrades is important before relying on Gemini in production.

Accomplishments

  1. Fully functional AI agent that works end-to-end — from text input to formatted report with DOCX export.

  2. Real-time data integration — Parallel Search actually searches the web for current data, not just returning pre-canned responses.

  3. Multi-location comparison — the ability to compare multiple filming locations in one report is genuinely useful for location scouting decisions.

  4. Graceful degradation — the app works perfectly in demo mode without any API keys, so it's always demoable.

  5. Ready for hackathon submission: Code is public on GitHub with MIT license, google-generativeai SDK integrated, Parallel API actively used at runtime, README complete.

What we learned

  1. AI agent development is iterative: Model APIs change frequently, requiring constant testing and adaptation. The model availability list is dynamic.

  2. Real-time data adds value: Users can immediately see the difference between demo mode (static data) and live mode (searched data). This validates the Parallel integration.

  3. Fallback strategies matter: Having a working demo mode when APIs fail is essential for hackathon presentations — you can't rely on external services being available.

  4. Prompt engineering for multi-section reports: Getting AI to consistently include comparison data across all sections requires explicit prompting and structured data input.

  5. Google Cloud billing is complex: Understanding the spending limits, credit requirements, and tier upgrades is important before relying on Gemini in production.

What's next

  1. Parallel Monitor integration — Set up real monitoring jobs that alert users when permit requirements or regulations change in their filming locations.

  2. More location data — Expand pre-loaded demo data to cover more countries and cities worldwide.

  3. Budget calculator — Interactive budget breakdown where users can adjust crew size, shoot days, and equipment to see real-time cost estimates.

  4. Crew/supplier marketplace — Integrate with vendor databases to allow direct contact or booking of local crew and equipment rental.

  5. Multi-language support — Currently English-only; adding Spanish, Portuguese, and French would expand reach to Latin American and European productions.

  6. Mobile app — A companion mobile app for location scouts to quickly look up permit requirements and contacts on location.

  7. Production templates — Pre-configured templates for common production types (commercial, music video, feature film, documentary) with typical crew counts and equipment needs.ion

Built With

Share this project:

Updates

Submission history