Page 595 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

REQ-07: The system shall integrate with relevant government databases for data verification
and cross-referencing.

REQ-08: The system shall ensure secure data handling and provide a user interface for
authorized access and investigation. 4.6: Finance

REQ-09: The underlying AI/ML models shall be adaptable and updated to maintain detection
accuracy.

3�1 Separate AI Models and Infrastructure
AI Models:

• Anomaly Detection: Trained on financial data using scikit-learn or TensorFlow for
detecting unusual transactions.
• Graph Analytics: Uses NetworkX to identify duplicate registrations via entity networks.
• NLP/Sentiment Analysis: Fine-tuned BERT models for analyzing social media/news.
• Geospatial Analysis: Custom algorithms to correlate satellite imagery with project
locations.
• OCR: Tesseract for document text extraction.
Infrastructure:

• Data Sources:
• Primary: NGO Darpan (ngodarpan.gov.in) for registration and financial data.
• Secondary: ISRO for satellite imagery, MCA for CSR data, social media APIs (e.g.,
Twitter/X).
• APIs:
• RESTful APIs for real-time data ingestion from NGO Darpan and MCA.
• ISRO Bhuvan API for geospatial data.
• Twitter/X API for sentiment analysis.
• Data Collection Protocols:
• Secure HTTPS for data transfer.
• Batch processing for historical data; streaming for real-time updates.
• SOPs:
• Data Ingestion: Validate data formats (CSV, JSON) and log errors.
• Model Training: Retrain models quarterly with new data; validate with 80/20 train-test
split.
• Alert Handling: Auditors review high-risk alerts within 48 hours; escalate critical cases to
NIC leadership.
• Privacy Compliance: Anonymize whistleblower data; encrypt financial records per India’s
PDP Act.

559