The problem Confidential AI Image Interpreter solves

Confidential Ai Image Interpreter leverages AI within Intel TDX and AMD SEV confidential VMs to provide a secure, privacy-preserving way to extract text from images and documents. Unlike traditional OCR services that may expose sensitive data to third-party servers, this solution ensures that all document processing occurs within a confidential computing environment. This makes it ideal for handling legal contracts, medical records, financial documents, and government files without the risk of unauthorized access. Businesses can securely digitize and analyze sensitive information, while researchers, lawyers, and journalists can process confidential materials without exposing them to external entities.

Web3 and blockchain applications can also benefit by integrating confidential OCR into smart contract workflows, ensuring data privacy in decentralized environments.

User Interaction and Data Flow

The user interacts with a chat-based AI interface on the frontend using Coinbase AgentKit , which is responsible for initiating the process. When a user requests image-based text extraction, the AI generates a ticket and provides an upload URL. This ticket is stored temporarily in memory, mapping it to a callback function that will process the uploaded image. The user then submits an image (JPEG, JPG, or PNG) via the provided upload endpoint, where the server validates the request using the ticket. Once the image is received, it is passed to PaddleOCR for text extraction.

After extracting the raw text, the system invokes a large language model (LLM) to interpret the OCR result and structure it into a meaningful JSON response. This JSON output includes an identity analysis, determining if the extracted text belongs to an ID document and parsing details such as country, name, and ID number. The AI then returns the structured data to the frontend, where the user can review the results. Throughout the process, all interactions occur within a confidential computing environment using Intel TDX and AMD SEV VMs, ensuring that no plaintext data is exposed outside the secure execution environment. The system also maintains a zero-trust architecture, preventing unauthorized access to sensitive documents while enabling secure and efficient OCR processing.

Product Integrations

• Chat interface with Secret AI agent (Intel TDX and AMD SEV confidential VMs)
• Integration with Coinbase CDP
• ID verification using PaddleOCR
• Secure file upload handling

Key differentiators and uniqueness of the project

Coinbase AgentKit with custom Secret AI LLM integration (Intel TDX and AMD SEV confidential VMs)