ProofLedger AI

• 

  ProofLedger AI dashboard showing HOLD, BLOCK, and HUMAN_REVIEW_ONLY receipts for AI-agent decisions.

Inspiration

AI agents can prepare business decisions quickly. But a recommendation is not the same as an approved action. A refund, claim, onboarding step, or account update may still be missing evidence, violate policy, or reuse the wrong case context.

What it does

ProofLedger AI is an evidence memory app for AI-agent decisions. It stores each recommendation as a receipt in AWS DynamoDB, checks evidence completeness and action boundaries, and shows a Vercel review dashboard. The result is HOLD, BLOCK, or HUMAN_REVIEW_ONLY. The app never executes the business action.

How we built it

We built a Next.js dashboard deployable on Vercel and designed the UI so it can be scaffolded and iterated with v0. The backend uses DynamoDB as an append-only receipt ledger. Each receipt includes a subject ID, decision status, evidence fields, policy findings, redaction status, and a receipt hash.

Challenges we ran into

The main challenge was keeping the action boundary clear. It is easy to make an AI workflow look complete. It is harder to keep a durable record of what evidence supports the recommendation and what action is still not allowed.

Accomplishments that we're proud of

We built a working full-stack prototype with Vercel and DynamoDB. It catches missing evidence, policy bypass, subject mismatch, unsafe external action, and public-safety redaction issues while keeping the strongest result limited to human review.

What we learned

The database is not just storage. For AI-agent systems, it becomes the memory of what was claimed, what evidence existed, and what action was allowed.

What's next for ProofLedger AI

Next we would connect ProofLedger to real enterprise workflow systems, add reviewer identity and approval attestations, and support customer-specific policy packs.

Built With

• amazon-dynamodb
• aws-sdk-for-javascript
• next.js
• react
• typescript
• v0
• vercel