CineNeuro — AI-Powered Audience Intelligence Platform

How It Works

Movie Trailer (MP4)
 |
 v
+-------------------------------+
| Multimodal Feature |
| Extraction |
| |
| Video -> V-JEPA2 (4.1GB) |
| Audio -> Wav2Vec-BERT |
| Text -> Llama 3.2 3B |
+-------------------------------+
 |
 v
+-------------------------------+
| TRIBE v2 Brain Encoding |
| (Meta FAIR) |
| |
| Predicts fMRI responses |
| across 20,484 vertices on |
| fsaverage5 cortical mesh |
+-------------------------------+
 |
 v
+-------------------------------+
| Emotion Mapping |
| |
| 7 brain regions mapped to |
| 5 emotion channels using |
| neuroscience-based weights |
+-------------------------------+
 |
 v
+-------------------------------+
| Intelligence Layer |
| |
| - Scene Detection |
| - Persona Simulation |
| - Competitive Benchmarking |
| - PDF Report Generation |
+-------------------------------+

Features

Engagement Timeline

Scene Intelligence

Audience Personas

Competitive Benchmarking

PDF Report

Tech Stack

Architecture

 +------------------+
 | React Frontend |
 | (Static Build) |
 +--------+---------+
 |
 +--------v---------+
 | Nginx |
 | (SSL Termination|
 | + Reverse Proxy)|
 +--------+---------+
 |
 +--------v---------+
 | FastAPI |
 | |
 | /api/v1/demos |
 | /api/v1/demo/:id|
 | /api/v1/analyze |
 | /reports/* |
 | /* (React SPA) |
 +--------+---------+
 |
 +--------------+--------------+
 | |
 +---------v----------+ +------------v-----------+
 | Pre-computed Results| | Live Pipeline (GPU) |
 | (JSON + PDF) | | |
 | 3 demo trailers | | preprocess -> TRIBE |
 +--------------------+ | -> emotions -> scenes |
 | -> personas -> bench |
 | -> PDF report |
 +-------------------------+

Project Structure

CineNeuro/
├── backend/
│ └── app/
│ ├── main.py # FastAPI app + static file serving
│ ├── config.py # Paths, model settings, constraints
│ ├── models/
│ │ └── schemas.py # 7 Pydantic models
│ ├── routers/
│ │ └── analyze.py # 5 API endpoints
│ └── services/
│ ├── video_ingestion.py # Upload validation + storage
│ ├── video_preprocessing.py # Multimodal feature extraction
│ ├── tribe_inference.py # TRIBE v2 brain prediction
│ ├── emotion_mapping.py # Brain vertices -> 5 emotions
│ ├── scene_intelligence.py # Peak/drop detection
│ ├── persona_simulation.py # 3 audience persona models
│ ├── benchmarking.py # Competitive comparison
│ └── report_generator.py # PDF report with ReportLab
├── frontend/
│ └── src/
│ ├── App.js # React dashboard
│ ├── App.css # Dark theme + animations
│ └── api.js # Axios API client
├── data/
│ └── results/ # Pre-computed JSON + PDF reports
├── scripts/
│ └── run_trailer.py # Standalone pipeline runner
├── Dockerfile # Multi-stage: Node build + Python serve
├── requirements.txt
└── .dockerignore

API Endpoints

Brain Region Mapping

Running Locally

Prerequisites

Development Mode

# Backend
cd CineNeuro
pip install -r requirements.txt
uvicorn backend.app.main:app --reload
# Frontend (separate terminal)
cd frontend
npm install
npm start

Docker

docker build -t cineneuro .
docker run -p 8000:8000 cineneuro
# Open http://localhost:8000

GPU Inference (for new trailers)

# On a GPU instance (g4dn.xlarge recommended)
pip install tribe-v2 neuralset whisperx torch
PYTHONPATH=. python scripts/run_trailer.py /path/to/trailer.mp4

Demo Trailers

Deployment

Built By