Adneural

Brain-Grounded Campaign Intelligence

The first platform that connects computational neuroscience to social simulation for pre-launch advertising analysis.

Demo Video · How It Works · Quick Start · Architecture · Research

---

The Problem

The global advertising industry spends $1 trillion annually. The neuromarketing industry — the science of understanding how brains actually respond to ads — gets $1.8 billion. That's 0.2%.

Why? A single fMRI study costs $50-200K, takes 6 months, and scans 20-30 people. By the time results arrive, the campaign has already launched.

Adneural changes that.

What We Built

Adneural is a three-engine platform that predicts how an advertisement will perform — neurologically and socially — before launch.

Engine	What It Does	Technology
Brain Encoding	Predicts cortical brain response to any video/audio/image	Meta TRIBE v2 (trained on 450h of fMRI data)
Video Summarizer	Generates scene-by-scene description of video content	LLM with frame extraction (vision-capable)
Social Simulation	Models how 200 personality-diverse AI agents react and interact	MiroFish/OASIS multi-agent framework
Diagnostic Agents	Analyzes brain + social data + video context, produces actionable reports	LangGraph + Neo4j GraphRAG over 22 papers

The novel contribution: the bridge between brain encoding and social simulation. We translate predicted cortical activation into personality-modulated emotional states that seed each agent's behavior. No one has connected these systems before.

Demo

Real-time cortical activation predicted by TRIBE v2 for the Jaguar "Copy Nothing" ad

Case Study: Jaguar "Copy Nothing" Rebrand

We ran the Jaguar rebrand ad through Adneural without knowing the real-world outcome. Our platform predicted:

Adneural Predicted	What Actually Happened
Weak emotional tagging, low memorability	Positive sentiment collapsed from 23% to 8%
No narrative processing (pure visual stimulus)	"Where are the cars?" became the meme
Narrative vacuum vulnerable to hostile reframing	The void where a car should have been got filled with ridicule
Minimal trust formation	No brand equity built. CEO resigned.

Launch Recommendation: NO GO

How It Works

Video Upload                     +-----------------------------+
     |                           | AI Video Summarizer         |
     |                           | (scene-by-scene description)|
TRIBE v2 (brain encoding)       | Added to every step below   |
     |                           +-----------------------------+
148 cortical regions,
second by second
     |
     +------------------+
     |                  |
+----v-----------+   +--v-------------------+
| Seeding Bridge |   | Neuro-Translator     |
|                |   |                      |
| 10 networks    |   | Neo4j GraphRAG       |
| -> 9D emotion  |   | 29 papers            |
| -> Big Five    |   | 2,684 findings       |
|   modulation   |   | 9,921 relationships  |
+-------+--------+   +---------+------------+
        |                       |
+-------v--------+              |
| MiroFish/OASIS |              |
| 200 AI agents  |              |
| Simulated      |              |
| Reddit         |              |
+-------+--------+              |
        |                       |
+-------v-----------+           |
| Social Analyst    <-----------+
| Cascade detection |
| Sentiment analysis|
+-------+-----------+
        |
        v
+-----------------------------------+
| Strategist Agent                  |
| Merges neural + social findings   |
| Computes risk scores              |
| Generates multi-audience reports  |
+-----------------+-----------------+
                  |
      +-----------v-------------+
      |   Guardrail Check       |
      |   (no manipulation)     |
      +-----------+-------------+
                  |
   +------+-------+-------+------------+
   v      v               v            v
Executive Marketing   Compliance   Full Research
 Report    Report      Report       Report

Quick Start

Prerequisites

• Python 3.10+
• Docker (for Neo4j)
• Node.js 18+ (for frontend)

Setup

# Clone
git clone https://github.com/OmarMusayev/neuro-social-platform.git
cd neuro-social-platform

# Environment
cp .env.example .env
# Edit .env with your API keys

# Install Python dependencies
pip install -r requirements.txt

# Start Neo4j
docker compose up -d

# Build the knowledge graph (first time only)
python -m rag.build_graph --papers-dir ./research/papers/

# Run analysis on the Jaguar demo
python -m scripts.run_analysis \
  --tribe-data ./data/demo/jaguar_tribe_output.json \
  --simulation-data ./data/demo/jaguar_simulation_output.json \
  --video ./path/to/jaguar_ad.mp4 \
  --stimulus "Jaguar Copy Nothing rebrand commercial" \
  --audience marketing \
  --output ./output/report.json

# Start the frontend
cd frontend && npm install && npm run dev

Environment Variables

# LLM (OpenRouter recommended)
LLM_PROVIDER=openrouter
LLM_API_KEY=your-key-here
LLM_MODEL=qwen/qwen3-235b-a22b
LLM_BASE_URL=https://openrouter.ai/api/v1

# Neo4j GraphRAG
NEO4J_RAG_URI=bolt://localhost:7688
NEO4J_RAG_USER=neo4j
NEO4J_RAG_PASSWORD=neurosocial123

The Seeding Bridge

The bridge is our novel contribution. It translates brain activation into agent emotional states:

1. Network Aggregation: 148 cortical parcels -> 8 functional brain networks
2. Emotional Computation: Network activations -> 9D emotional state vector (anxiety, trust, excitement, discomfort, memorability, etc.)
3. Personality Modulation: Base emotional state x Big Five personality traits -> individualized response. A neurotic viewer gets anxiety. An extraverted viewer gets curiosity. Same brain data, different reactions.
4. Population Distribution: 200 agents across 6 archetypes (enthusiastic sharer, empathetic worrier, skeptical analyst, hostile critic, average viewer, emotional storyteller)

Research

Our brain-to-emotion mapping is grounded in peer-reviewed neuroscience literature:

• Corbetta & Shulman (2002) — Attention networks
• Kanwisher et al. (1997) — Face processing
• Klucharev et al. (2008) — Persuasion and memory encoding
• Genevsky et al. (2025) — Neuroforecasting
• Full list of 22 papers

Subcortical Inference

TRIBE v2 outputs cortical surface data (~20,484 vertices). Subcortical structures (amygdala, hippocampus, nucleus accumbens) are inferred from cortical proxy patterns with explicit confidence tagging, not directly measured. This is documented honestly throughout our codebase and reports.

Architecture

Component	Technology	Purpose
Brain Encoding	Meta TRIBE v2	Predict fMRI-like cortical activation from media
Video Summarizer	LLM + ffmpeg frame extraction	AI-generated scene-by-scene video description
Atlas Mapping	Destrieux (aparc.a2009s)	Aggregate vertices into named brain regions
Knowledge Graph	Neo4j + sentence-transformers	GraphRAG over 22 neuroscience papers
Emotional Bridge	Custom Python	Cortical activation -> 9D emotional state
Social Simulation	MiroFish/OASIS	Multi-agent social media simulation
Diagnostic Agents	LangGraph	Three-agent analysis pipeline + video context
LLM Backend	OpenRouter (Qwen 3)	Model-agnostic, swappable via env vars
Frontend	Next.js 15 / React 19 / Three.js	Dashboard and demo interface

Project Structure

neuro-social-platform/
├── tribe/          # TRIBE v2 brain encoding integration
├── bridge/         # Brain data -> emotional seeding bridge
├── simulation/     # MiroFish/OASIS simulation adapters
├── rag/            # Neo4j GraphRAG pipeline
├── agents/         # LangGraph diagnostic agents
├── api/            # FastAPI backend
├── frontend/       # Next.js 15 dashboard (React, Three.js, Tailwind)
├── scripts/        # CLI tools and utility scripts
├── data/           # Brain architecture + demo data
├── research/       # Analysis outputs and paper references
├── cluster/        # HPC cluster scripts (TRIBE v2)
├── demo_visuals/   # Auto-animated visual assets
├── docs/           # Architecture and pipeline documentation
└── tests/          # Integration tests

Reports

Adneural generates four audience-specific reports from the same analysis:

Report	Audience	Focus
Executive	C-suite	Go/no-go recommendation with risk dashboard
Marketing	Creative team	Timestamped edit recommendations with neural rationale
Compliance	Legal/ethics	Mental health impact assessment, regulatory flags
Full Research	Technical	Complete neural-social bridge analysis with paper citations

Team

Built at Purdue Catapult Hackathon by:

• Omar Musayev
• Pranjal Bhatia
• Ved Karamnchandi
• Aadit Kedia
• Piyush Dua
• Rishab Shenai

License

MIT — see LICENSE for details.

Acknowledgments

• Meta FAIR for TRIBE v2
• CAMEL-AI for the OASIS framework
• MiroFish for the social simulation engine
• The authors of the 22 neuroscience papers that form our knowledge base