DocuTrust Africa

Inspiration

In Central and West Africa, paper-based administrative systems are highly vulnerable. We've seen firsthand how forged birth certificates, counterfeit land titles, and fake academic degrees lock legitimate citizens out of the financial system and breed mistrust in institutions. The recent explosion of Generative AI has made things worse, allowing malicious actors to create hyper-realistic fake documents and deepfakes. We realized that traditional OCR and manual checks were no longer enough. We needed a system that doesn't just "read" a document, but cryptographically proves its authenticity and human origin.

What it does

DocuTrust Africa is a hybrid AI and Web3 API. When a government agency or a bank receives a document or media file, our system passes it through a 3-tier pipeline:

Structural Analysis: Checks formatting, stamps, and watermarks against official templates.
Digital Forensics: Analyzes metadata, compression artifacts, and pixel inconsistencies to detect manual tampering (e.g., splicing).
AI-Generation Detection: Identifies the statistical signatures of deepfakes and AI-generated content. Once validated, the file's unique SHA-256 hash is anchored to the blockchain, creating an immutable, time-stamped proof of authenticity without exposing the sensitive data itself.

How we built it

We designed a decoupled, microservices-based architecture to ensure scalability:

The Brain (AI Backend): We built a RESTful API using FastAPI (Python). We integrated computer vision libraries (OpenCV) for forensic analysis and fine-tuned pre-trained CNN models via PyTorch to detect AI-generated artifacts.
The Ledger (Blockchain): We wrote Smart Contracts in Solidity and deployed them on the Polygon network to ensure fast transaction times and near-zero gas fees. We used Ethers.js to bridge our backend with the blockchain.
The Interface (Frontend): We created a clean, intuitive dashboard using React and TailwindCSS, allowing non-technical administrative staff to easily drag-and-drop files and instantly receive a Trust Score and a Blockchain transaction receipt.

Challenges we ran into

Our biggest challenge was the AI evaluation metric trade-off. In digital identity, a False Acceptance (letting a deepfake pass) is far more dangerous than a False Rejection (rejecting a blurry but real document). We had to heavily tweak our confidence thresholds to prioritize a near-zero False Acceptance Rate (FAR). Another challenge was data privacy: we couldn't store legal documents on a public ledger. We solved this by only hashing the documents ($H(document) = hash$) and anchoring the hash, ensuring full compliance with data protection laws.

What we learned

We learned how to bridge Web2 infrastructure with Web3 trust mechanisms. We also deepened our understanding of digital forensics—specifically how Generative AI leaves microscopic traces in frequency domains that the human eye cannot see, but that AI can be trained to catch.

What's next for DocuTrust Africa

Our next step is to partner with a local university or a microfinance institution to pilot the API. We also plan to develop a mobile SDK (React Native/Flutter) so citizens can verify documents directly from their smartphones by scanning a secure QR code linked to the blockchain hash.