🌐 CineIntel Engine

Real-Time Web Grounding, Film Trend Research & Autonomous Script Doctor

Built for the Parallel Track — Google Cloud "Agentic Cinema: The Blockbuster Hackathon"

---

🌟 Overview

CineIntel Engine is an autonomous film intelligence and research platform featuring implemented live integrations of the official Parallel Search API / SDK (parallel-web) and Google Gemini 3.7 Flash. By default, the public app provides deterministic Parallel-shaped research fixtures alongside a local Gemini-shaped synthesis fixture.

The public evaluator runs in live mode: autonomous agents query Parallel Search and preserve titles, URLs, excerpts, and retrieval status while Gemini 3.7 Flash runs through Vertex AI. Public end-to-end verification returned live web sources through the stable judge URL.

---

🚀 Key Features

1. 🔍 Direct Parallel Web Search Integration:
   • Implements the official parallel-web>=0.5.0 Python SDK (from parallel import AsyncParallel) and x-api-key request structure (objective and search_queries) to retrieve web research and excerpts in live mode.
2. ✍️ Fact-Grounded Screenplay Architecture:
   • When configured in live mode, ingests real-world excerpts and visual benchmarks into Gemini 3.7 Flash screenplay treatments.
3. 🎭 Audience Archetype & Casting Intelligence:
   • Evaluates character archetype resonance against genre benchmarks and voice profile recommendations.
4. 🤖 Specialized Gemini 3.7 Agent Crew:
   • TrendAnalystAgent: Researches open-web narrative tropes and presents evidence with uncertainty factors.
   • GroundedWriterAgent: Synthesizes Parallel search excerpts into grounded screenplay treatments and scene beats.
   • CastingIntelAgent: Evaluates character archetypes and casting considerations.
5. 🛡️ Honest Evidence & Strict Live Error Handling:
   • The default public app uses self-contained local research fixtures with non-clickable fixture IDs (fixture:...).
   • In configured live mode, failures return explicit live_unavailable/live_error without silent demo fallback.

---

⚡ Quickstart

cd backend
pip install -r requirements.txt
python run_backend.py

API docs available at http://localhost:8002/docs and Web UI at http://localhost:8002/.

---

🧪 Testing

cd backend
python -m pytest -v

---

☁️ Vercel Deployment

Live judge demo: https://cineintel-engine.vercel.app/
Public source: https://github.com/AtchayamG/cineintel-engine

The stable Vercel judge URL proxies Cloud Run revision cineintel-engine-live-00003-6gt. Public health reports runtime_mode: live; a public trend-analysis call returned URL-bearing sources with retrieval_status: RETRIEVED_LIVE_WEB and parallel_mode: live. See docs/evidence/PUBLIC_RUNTIME_VERIFICATION_2026-08-30.md.

When deploying to Vercel, ensure you set the Root Directory in your Vercel project settings to backend.

# Install the Vercel CLI
npm i -g vercel

# Deploy the project from the root directory
vercel

Note: To run in live mode, ensure you set both PARALLEL_API_KEY and GEMINI_API_KEY environment variables in your Vercel project settings and change RUNTIME_MODE to live.

---

📄 License

Licensed under the MIT License.