ogulcanaydogan/AI-Provenance-Tracker
Open-source multi-modal AI content detection platform, analyses text, images, audio, and video for AI provenance with explainable scoring, multi-provider consensus, public benchmarks, and enterprise deployment (Helm, Terraform)
What's novel
Open-source multi-modal AI content detection platform, analyses text, images, audio, and video for AI provenance with explainable scoring, multi-provider consensus, public benchmarks, and enterprise deployment (Helm, Terraform)
Code Analysis
20 files read · 6 roundsA multi-modal AI-generated content detection platform that combines local heuristic/ML detectors with external provider consensus voting (Copyleaks, Reality Defender, Hive, C2PA) to produce explainable evidence cards for text, image, audio, and video authenticity analysis.
Strengths
The text detection pipeline is genuinely sophisticated with domain-aware calibration, Platt scaling, chunk consistency analysis, and graceful ML fallback. The provider consensus architecture with weighted voting, disagreement detection, and C2PA content credentials verification is well-designed and production-ready.
Weaknesses
Image, audio, and video detectors are explicitly MVP-level statistical heuristics (FFT, spectral flatness, color banding) rather than trained models, limiting their real-world accuracy. The project is large in scope (689 files) but the core detection depth is concentrated almost entirely in the text modality.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
146
Contributors
2
Files
792
Active weeks
5
Repository
Language
Python
Stars
2
Forks
1
License
MIT