Explainable AI Platform for Detecting Forged or Manipulated Documents
Forensica AI is a cybersecurity-focused forensic analysis platform that detects potential document tampering using a modular AI-assisted pipeline. The system analyzes uploaded documents through multiple forensic stages including OCR extraction, anomaly detection, layout analysis, font verification, and metadata inspection to determine authenticity.
This project was developed for ThinkRoot x Vortex ’26 Hackathon — Track C.
PROBLEM
Digital documents such as certificates, transcripts, invoices, and identity records are frequently manipulated using image editing software. Traditional verification methods rely on manual inspection which is slow, error-prone, and requires specialized expertise.
There is a need for a scalable automated system that assists investigators, institutions, and organizations in detecting document forgery quickly while providing explainable evidence.
SOLUTION
Forensica AI provides a modular forensic analysis pipeline capable of automatically analyzing documents and highlighting suspicious regions.
The system:
• Extracts text using OCR
• Detects visual manipulation artifacts
• Identifies font inconsistencies
• Evaluates layout structure
• Analyzes metadata
• Aggregates risk signals into an interpretable forensic score
• Generates a structured verification report
The goal is not just detection but explainable forensic insights, allowing investigators to understand why a document is considered suspicious.
KEY FEATURES
• Automated document forgery detection
• Explainable anomaly detection
• OCR-based text extraction
• Pixel anomaly and cloning detection
• Font inconsistency detection
• Layout integrity analysis
• Metadata forensics
• Risk scoring engine
• Secure per-user document history
• Batch document analysis
SYSTEM ARCHITECTURE
Frontend Dashboard
↓
FastAPI Backend API
↓
Forensic Analysis Pipeline
• File Preprocessor
• OCR Extraction
• Pixel & Anomaly Detection
• Layout Consistency Check
• Font Detection
• Metadata Forensics
• Risk Scoring Engine
↓
Result Aggregation + Forensic Report
FORENSIC ANALYSIS PIPELINE
-
File Preprocessing
Image normalization, noise filtering, and format standardization. -
OCR Extraction
Text is extracted using Tesseract OCR and analyzed for structural inconsistencies. -
Pixel Anomaly Detection
Detects cloned regions, editing artifacts, and unnatural pixel patterns. -
Layout Integrity Check
Analyzes alignment, spacing, and structural layout inconsistencies. -
Font Detection
Detects mismatched fonts and typographic anomalies often introduced by edits. -
Metadata Forensics
Extracts embedded metadata and identifies suspicious editing timestamps. -
Risk Scoring Engine
All signals are aggregated to produce a forgery probability score.
TECHNOLOGY STACK
Backend • Python • FastAPI • OpenCV • Tesseract OCR • NumPy
Frontend • HTML • CSS • JavaScript
Infrastructure • Firebase Authentication • Firebase Firestore • Firebase Storage • Docker • Netlify
STORAGE STRATEGY
Cloud Mode
Firebase is used for:
• User authentication
• Secure document storage
• Analysis result persistence
Local Mode
For development and resource efficiency, the system can run locally with temporary storage and offline analysis.
REPOSITORY STRUCTURE
backend/ models/ routes/ services/ utils/
frontend/ css/ js/
RUNNING THE PROJECT LOCALLY
Clone the repository
git clone https://github.com/NovocaineX/vortex-hackathon26.git
Install dependencies
pip install -r requirements.txt
Run the backend server
python backend/main.py
Open the frontend in your browser and upload a document to start analysis.
SECURITY AND PRIVACY
• Each user's uploads are isolated
• Documents are encrypted in storage
• Access control enforced through Firebase authentication
• Analysis history is restricted per user
FUTURE IMPROVEMENTS
• Deep learning based forgery detection models
• Blockchain-backed document verification
• Institutional verification APIs
• Multilingual OCR support
• Large scale batch verification
DEMO
Demo video and screenshots are available in the demo directory.
AUTHOR
Developed by Aadarsh (NovocaineX) for ThinkRoot x Vortex ’26 Hackathon.
