Using Granite 3.2 Vision in a TEE for Self-Custody KYC:

Hi all,

I'm exploring IBM's Granite 3.2 Vision model to extract and hash personal information from documents (e.g., IDs, passports) within a Trusted Execution Environment (TEE) for a self-custody Know Your Customer (KYC) system. The goal is to securely process sensitive data, hash it for verification, and ensure compliance while maintaining user privacy.

Problem: Personal information restrictions are blocking progress due to LLM personal information restrictions. Running Granite 3.2 Vision in a TEE seems promising for secure execution, but I'm hitting roadblocks with the LLM not allowing this interaction.

What I'm looking for:

1. Advice on configuring Granite 3.2 Vision in a TEE to extract personal data (e.g., text from ID images).

2. Workarounds for compliance with data protection laws while still enabling self-custody KYC verification.

3. Examples or alternative tools for similar use cases.

Has anyone tackled this or similar challenges? Any tips on model setup? Open to creative solutions or pointers to relevant projects.

Thanks in advance!

-Braydn Larsen