AI-powered social intelligence platform for detecting, correlating, and investigating emerging threats across fragmented social signals.
INTOCIAL is an end-to-end social intelligence platform that combines multi-source signal ingestion, AI-assisted risk analysis, incident management, cross-signal correlation, multimodal image analysis, analytics, and intelligence reporting into a unified operational dashboard.
Modern social platforms generate large volumes of fragmented and rapidly changing information. INTOCIAL is designed to transform these signals into structured intelligence that can be reviewed, correlated, and investigated from a single workspace.
- Multi-source social signal ingestion
- AI-assisted risk and threat analysis
- Entity and contextual analysis
- Explainable risk scoring
- Incident creation and management
- Cross-signal and cross-platform correlation
- Temporal and event-level correlation
- Risk alerts and acknowledgement workflows
- Image-based multimodal analysis
- Historical analytics and anomaly detection
- Intelligence report generation
- Authenticated API access
- Operational investigation dashboard
INTOCIAL uses a common signal-processing architecture for multiple social platforms.
| Source | Current Status |
|---|---|
| YouTube | Real API ingestion |
| Simulated / mock ingestion | |
| X | Simulated / mock ingestion |
| Simulated / mock ingestion |
YouTube is currently the primary real external-data ingestion source. The other connectors use simulated data for development and demonstration.
Signals are processed through an analytical pipeline containing components for:
- NLP processing
- Entity extraction
- Contextual analysis
- Risk scoring
- Threat indicators
- Explainable risk components
- Anomaly detection
- Signal-level intelligence
The system is designed to expose analytical evidence behind risk assessments rather than relying only on an opaque prediction.
Related signals can be grouped into incidents based on similarity and contextual relationships.
The incident workflow supports:
- Incident creation
- Signal attachment
- Incident risk calculation
- Similarity-based grouping
- Signal aggregation
- Incident-level intelligence
- Incident management
INTOCIAL correlates signals across multiple dimensions:
- Content similarity
- Entity overlap
- Cross-platform corroboration
- Temporal relationships
- Event-level relationships
This helps transform fragmented signals into connected investigative evidence.
The alerting layer provides:
- Risk-based alerts
- Alert severity
- Alert details
- Alert acknowledgement
- Connection between analytical risk and operational review
INTOCIAL currently supports image-based multimodal analysis.
The system can analyze image inputs as part of the intelligence workflow.
Video analysis is not currently implemented.
The analytics workspace provides:
- Historical risk trends
- Platform distribution
- Risk distribution
- High-risk activity
- Anomaly detection
- Historical signal analysis
Investigations and incidents can be converted into structured intelligence reports containing analytical evidence and explanations for further review.
+---------------------+
| Social Sources |
| YouTube / Reddit / |
| X / Instagram |
+----------+----------+
|
v
+---------------------+
| Signal Ingestion |
| & Normalization |
+----------+----------+
|
v
+---------------------+
| AI Analysis |
| NLP / Entities / |
| Context / Risk |
+----------+----------+
|
+----------------+----------------+
| | |
v v v
+------------+ +------------+ +------------+
| Incidents | | Correlation| | Alerts |
+-----+------+ +------+-----+ +------+-----+
| | |
+-----------------+----------------+
|
v
+---------------------+
| Intelligence Layer |
| Reports / Analytics |
+----------+----------+
|
v
+---------------------+
| INTOCIAL Dashboard |
| Investigation UI |
+---------------------+
- Python
- FastAPI
- Pydantic
- SQLAlchemy
- PostgreSQL
- psycopg2
- JWT authentication
- PyTorch
- Hugging Face Transformers
- scikit-learn
- spaCy
- NumPy
- Pandas
- Pillow
- React
- Vite
- React Router
- Axios
- Recharts
- Lucide React
- Pytest
- Jupyter
- Conda
- Git
social-media-analytics/ | +-- dashboard/ | +-- public/ | +-- src/ | +-- components/ | +-- pages/ | +-- services/ | +-- package.json | +-- vite.config.js | +-- data/ | +-- raw/ | +-- processed/ | +-- live/ | +-- models/ | +-- distilbert-crisisbench/ | +-- risk_config.json | +-- notebooks/ | +-- src/ | +-- api/ | +-- data/ | +-- database/ | +-- models/ | +-- services/ | +-- tests/ | +-- migrate_signals.py +-- simulate_stream.py +-- requirements.txt +-- .env.example +-- .gitignore +-- README.md
Before running INTOCIAL locally, install:
- Python 3.x
- Node.js
- PostgreSQL
- Git
- Conda or another Python environment manager
git clone https://github.com/ridamgupta79-dev/social-media-analytics.git cd social-media-analytics
conda create -n social-media-analytics python=3.12 conda activate social-media-analytics
pip install -r requirements.txt
python -m spacy download en_core_web_sm
Create a local .env file from the provided template.
On Windows PowerShell:
Copy-Item .env.example .env
Configure the required credentials and database connection inside .env.
The project uses the following environment variables:
X_API_KEY= X_API_SECRET= X_ACCESS_TOKEN= X_ACCESS_TOKEN_SECRET=
REDDIT_CLIENT_ID= REDDIT_CLIENT_SECRET= REDDIT_USER_AGENT=
YOUTUBE_API_KEY=
INSTAGRAM_ACCESS_TOKEN=
DATABASE_URL=
SECRET_KEY=
Never commit .env or real API credentials to GitHub.
INTOCIAL uses PostgreSQL for persistent application data.
Configure the PostgreSQL connection through:
DATABASE_URL=
The project includes database initialization and migration-related functionality.
From the project root:
uvicorn src.api.main:app --reload
Backend:
FastAPI documentation:
Open a second terminal:
cd dashboard npm install npm run dev
Frontend:
The backend provides functionality across areas including:
/auth /predict /analyze /analyze-multimodal /ingest /ingest/youtube /ingest/reddit /ingest/x /ingest/instagram /incidents /alerts /correlation /reports /analytics /health
Interactive API documentation is available through FastAPI Swagger UI at:
The application includes security measures such as:
- JWT-based authentication
- Protected intelligence endpoints
- Authenticated frontend API requests
- Restricted development CORS configuration
- Environment-based secret management
- Image upload validation
- File-size restrictions for image uploads
- Sensitive configuration excluded through .gitignore
Parts of the machine-learning workflow use the CrisisMMD / CrisisBench dataset for model development and experimentation.
The dataset is not included in this repository.
Large raw datasets, processed datasets, and generated data are excluded from GitHub.
To reproduce the relevant ML experiments, obtain the required dataset separately and place it in the expected local data directories.
Large trained model weights and training checkpoints are intentionally excluded from GitHub.
The repository keeps lightweight configuration and tokenizer files where useful, while large model artifacts such as:
*.safetensors *.pt *.pth *.pkl
are excluded.
This keeps the repository manageable while preserving the source code and model configuration.
Run the test suite with:
pytest
The project contains tests covering major application workflows including:
- Authentication
- Risk analysis
- Incidents
- Alerts
- Correlation
- Intelligence workflows
- API functionality
INTOCIAL is currently a local research and prototype platform.
The implemented system has been tested across:
- Frontend workflows
- Backend API workflows
- Authentication and session handling
- Signal ingestion
- Risk analysis
- Incident workflows
- Correlation workflows
- Alert workflows
- Multimodal image analysis
- Analytics
- Intelligence reporting
The YouTube connector is currently the primary real external-data ingestion path.
Reddit, X, and Instagram connectors currently use simulated data for development and demonstration.
Current limitations include:
- Reddit ingestion is simulated.
- X ingestion is simulated.
- Instagram ingestion is simulated.
- Multimodal analysis currently focuses on images.
- External API availability depends on provider credentials and quotas.
- Large ML model weights are not distributed with the repository.
- The current system is intended for local development and demonstration rather than production deployment.
Potential future development includes:
- Production-grade social platform integrations
- Distributed ingestion workers
- Real-time streaming infrastructure
- Scalable intelligence processing
- Advanced multimodal models
- Entity knowledge graphs
- Analyst collaboration workflows
- Production deployment
- Expanded automated evaluation and monitoring
INTOCIAL is designed as an intelligence analysis and decision-support platform.
Risk scores, correlations, alerts, and analytical outputs should be treated as investigative signals requiring appropriate human review and contextual verification rather than as automatically established facts.
Ridam Gupta
AI / Machine Learning · Software Development · Data & Intelligence Systems
INTOCIAL — Intelligence Operations & Social Intelligence Analysis
An AI/ML engineering project combining machine learning, data processing, backend systems, correlation intelligence, and an operational investigation interface.
