VeriPaper is a forensic-grade, multi-module verification platform designed to ensure the authenticity and integrity of academic research papers. It serves as a comprehensive, free alternative to commercial tools like Turnitin, providing deep insights into AI generation, plagiarism, and methodological consistency.
VeriPaper analyzes documents through six specialized forensic dimensions:
- AI Detection: Identifies synthetic text using a fine-tuned DistilBERT engine, optimized via int8 quantization for high-speed, low-memory inference.
- Zero-Cost Plagiarism Attribution: Detects overlaps with the global web corpus using an intelligent DuckDuckGo-based attribution engine—no expensive API keys required.
- Intent-Based Taxonomy: Automatically distinguishes between "Cited" and "Uncited" similarities, helping researchers identify missing attributions vs. legitimate citations.
- Citation Validation: Cross-references every DOI and reference against the CrossRef global database to detect "hallucinated" or retracted citations.
- Statistical Integrity: Analyzes P-value distributions and statistical patterns to identify potential data manipulation or reporting anomalies.
- Professional Reporting: Generates industry-standard PDF reports with executive summaries, credibility badges, and interactive full-text overlays.
VeriPaper is built for maximum efficiency on constrained environments:
- Engine: FastAPI + React 18.
- ML Stack: ONNX Runtime + standalone Rust tokenizers (optimized for 512MB RAM).
- Monitoring: Integrated Sentry error tracking and automated health monitoring.
- Deployment: Fully containerized (Docker) and optimized for Render's free tier.
For a deep dive into the system design, see ARCHITECTURE.md.
- Docker & Docker Compose
- Python 3.11+
- Node.js 20+
docker-compose up --buildThe platform will be available at http://localhost:8000.
Refer to CONTRIBUTING.md for detailed local development instructions.
- Phase 1: Core Analysis Infrastructure (AI, Plagiarism, Citations).
- Phase 2: Professional Reporting Engine (PDF, Overlays, Taxonomy).
- Phase 3: PCV Suite (Methodological Fingerprinting & Anomaly Detection).
- Phase 4: Visual Forensic Analysis (Image manipulation detection).
- Phase 5: Multi-language support for global research standards.
Distributed under the MIT License. See LICENSE for more information.
Disclaimer: VeriPaper is an automated assistant. Final integrity decisions should always be made by qualified human reviewers.