BillGaurd - Forensic Risk Assessment of Indian Financial Documents
- Tech Stack: Python, PyTorch, React, FastAPI, PostgreSQL, XGBoost, OpenCV, PyMuPDF
- Repository Access: Request Access via Email
- Live Demo: billgaurd.yugagarwal.dev
BillGaurd is a calibrated cross-layer forensic risk-assessment framework specifically designed for Indian financial documents to detect document tampering, financial fraud, and synthetic invoice generation.
Unlike systems relying on a single text source or image pixels, BillGaurd simultaneously analyzes visual, structural, textual, semantic, arithmetic, regulatory, identity, and provenance information.
- Field-Level Evidence Alignment - Identifies exactly which field (total amount, GSTIN, invoice number) is suspicious and why.
- OCR–Native PDF Text Comparison - Extracts both visible text and embedded PDF text separately to detect hidden text layers and overwritten values.
- Arithmetic-Consistent Tampering Detection - Validates mathematical relationships (quantity × price, tax calculations) to catch smart forgeries.
- Missing-Modality-Aware Evidence Fusion - Dynamically adapts based on available data, using learned models (neural networks, graph-based models) to generate a calibrated risk score.
- Indian GST-Aware Validation - Includes verification of GSTIN structure, state codes, CGST/SGST/IGST rules, and HSN/SAC classification.
Designed as a robust decision-support tool, BillGaurd produces a highly explainable and transparent report listing high-risk findings, moderate anomalies, and authenticity indicators.