💡 Inspiration

In major film productions worldwide—especially Indian and Hollywood blockbuster shoots—a single day of outdoor shoot cancellation due to sudden unpredicted weather or road slickness costs between $50,000 to over $500,000 (₹40 Lakhs to ₹4 Crores) in equipment rental fees, lead actor dates, and crew downtime.

Worse, when heavy vehicle stunts or massive 50-foot Technocranes operate on damp asphalt or saturated mud, hydroplaning accidents and outrigger tip-overs cause catastrophic injuries.

Existing production tools are static Excel call sheets and generic weather phone apps that don't calculate real physics or coordinate autonomous agent crews. We built CrisisShift OS to give film directors an autonomous multi-agent command system powered by Gemini 3.8, grounding directorial decisions in real-time satellite atmospheric radar, surface asphalt friction equations ((\mu)), and instant bilingual Tanglish/English set dispatches.


🚀 What It Does

  1. Real-Time Data & Sensor Integration:

    • Connects live to OpenStreetMap Nominatim and Open-Meteo atmospheric radar to extract live precipitation, humidity, wind vectors, and temperatures for any shooting location on Earth.
    • Generates authentic ArcGIS high-resolution satellite optical reconnaissance and interactive Pydeck 3D GPS radar with a bright red live location beacon.
  2. Terrain Physics & Rig Safety Engine:

    • Computes dynamic asphalt friction coefficients ((\mu = 0.32) wet hydroplane hazard to (\mu = 0.85) optimal dry asphalt).
    • Evaluates soil density and bearing pressure capacity ((\sigma_{\text{max}} = 150 \text{ kPa})) to ensure heavy Technocranes don't sink or overturn.
  3. Gemini 3.8 Multi-Agent Directorial Swarm:

    • Agent 1 (Satellite Scout): Scans orbital radar sensors and atmospheric threat vectors.
    • Agent 2 (Terrain Physics Engine): Computes ground friction, braking distance multipliers, and crane base load.
    • Agent 3 (Director Swarm): Synthesizes live telemetry and makes deterministic calls: 🟢 FULL GREEN LIGHT vs 🔴 WEATHER HALT with indoor soundstage redirection.
    • Agent 4 (Crew Dispatcher): Formats emergency call sheets and WhatsApp broadcasts.
  4. Tanglish & Regional Indian Cinema Localization:

    • On shooting sets in South India and Bollywood, instructions are given in mixed regional dialects like Tanglish (Tamil + English). CrisisShift OS natively generates production dispatches in Tanglish, Tamil, Hindi, Telugu, and English so crew leads immediately grasp safety directives without translation friction.
  5. Mission Audit Database & PDF Export:

    • Automatically archives every recon run in an SQLite database.
    • Generates 1-click printable PDF Production Dispatches and Markdown reports for studio executives and insurance logs.

🏗️ How We Built It

CrisisShift OS is engineered as an enterprise-grade 3-Tier Architecture:

  • Frontend UI Layer (frontend/): Streamlit, Next-Gen Cyberpunk / Dark Obsidian & Electric Amber HUD Design System, Pydeck 3D interactive mapping, and custom typography.
  • Backend Services Layer (backend/): Google GenAI SDK (gemini-3.8-flash), Open-Meteo REST API, OpenStreetMap Nominatim Geocoding, ArcGIS World Imagery Server, Physics Engine (telemetry_service.py), and FPDF2 Unicode PDF Exporter (pdf_exporter.py).
  • Database Layer (database/): SQLite (crisis_shift.db) with schema for mission audit logs, scene breakdowns, and crew emergency rosters.

🧗 Challenges We Ran Into

  1. FPDF Unicode Emoji Exceptions:
    • Generating PDF dispatches failed initially with FPDFUnicodeEncodingException because core Latin fonts do not support 4-byte unicode emojis. We engineered a dedicated sanitize_pdf_text() pipeline that translates unicode emojis into clean, high-contrast text badges ([SATELLITE], [HALT], [GREEN LIGHT]) without crashing.
  2. Deterministic Physics Arbitration in Multi-Agent Swarm:
    • Ensuring that Gemini 3.8 never hallucinates a "Green Light" during torrential rain required grounding the prompt in verified mathematical thresholds (Rain > 1.0mm (\implies \mu < 0.40 \implies \text{HALT})).
  3. Browser Autofill Suppression:
    • Modern Chromium browsers aggressively displayed "Saved Info" dropdowns over coordinate inputs. We injected custom DOM attribute overrides (autocomplete="off", data-lpignore="true") to provide an uninterrupted manual typing experience.

🏆 Accomplishments That We're Proud Of

  • Sub-15 Second Autonomous Latency: Instant atmospheric recon, friction calculation, and directorial blueprint generation in under 15 seconds.
  • True Multi-Agent Determinism: Ground truth satellite data directly controls the agent swarm verdict—eliminating expensive production guesswork.
  • Set-Ready Tanglish Localization: Breaking language barriers on regional film sets by generating native Tanglish dispatches that First ADs can send directly to crew WhatsApp groups.
  • Exportable Audit Trail: Full SQLite database logging and instant PDF/MD dispatch downloads for production insurance compliance.

🎓 What We Learned

  • How to combine real-time physical telemetry models with LLM reasoning agents to enforce strict deterministic safety protocols rather than probabilistic guesses.
  • How to design high-density tactical HUD interfaces in Streamlit that feel like mission-critical aerospace / film production gear.
  • The nuances of multilingual set coordination in high-stress film production environments.

🔮 What's Next for CrisisShift OS

  • Managed MCP Server Adapter: Connecting CrisisShift OS via Model Context Protocol (MCP) directly to studio production accounting software (e.g., Movie Magic Scheduling) and real-time drone telemetry feeds.
  • Live Cast/Crew WhatsApp SMS Bot: Direct Twilio/WhatsApp Business API integration to dispatch emergency alerts to 200+ crew members simultaneously.

Built With

  • esri
  • fpdf2
  • gemini-3.8-flash
  • google-genai
  • multi-agent
  • open-meteo
  • pydeck
  • python
  • sqlite
  • streamlit
Share this project:

Updates

Submission history