Inspiration
A successful PayPal payment confirms what a customer ordered, but it does not confirm whether the correct products and quantities are actually packed before a parcel is sealed.
We wanted to explore this gap in physical order fulfilment. Small merchants and teams packing orders manually may need a better way to compare what the buyer paid for with what is being prepared for dispatch.
This led us to build ParcelProof AI, a PayPal-linked packing review prototype that brings payment information, product references and packing evidence into one workflow.
What it does
ParcelProof AI helps merchants review packed items against a PayPal-paid order before sealing the parcel.
The workflow starts with a PayPal Sandbox checkout. The backend verifies the completed order and payment capture, including the amount and currency. The purchased products and quantities become the source for a packing manifest.
The merchant confirms the packing list and finalizes it. ParcelProof creates a SHA-256 fingerprint of the declared contents.
Before sealing, the merchant uploads a photograph of the packed items. Google Gemini Vision compares the visible contents with the finalized manifest and saved product reference information.
The AI provides observations, while the merchant makes the final decision: approve, reject or request a correction. A correction decision keeps the shipment on hold and is recorded separately from the AI result.
ParcelProof also links the payment references, manifest, packing-photo hashes, AI observations and merchant decision in an application-level evidence history.
How we built it
We built the application using:
- PayPal Developer Platform – Orders v2 Sandbox: Checkout, order approval and payment capture verification.
- Google Gemini Vision: Visual inspection of packing photographs against the order-derived manifest and product references.
- Python and FastAPI: Backend APIs, payment verification, manifest creation, photo validation, AI inspection and review records.
- React and Vite: Merchant-facing interface for orders, product catalogue, packing and inspection review.
- Supabase PostgreSQL and private Storage: Hosted application records and evidence storage.
- Docker and Render: Application packaging and deployment.
PayPal is central to the workflow because the verified Sandbox order supplies the purchased items and quantities used for packing verification. Gemini then assists with reviewing the photographed items.
Challenges we ran into
One of our biggest challenges was handling uncertainty in AI-based visual inspection. Similar-looking products and catalogue variants can be difficult to distinguish from a photograph.
We also encountered AI processing delays and temporary provider failures. We worked on error handling, inspection history and clearer separation between failed attempts, AI observations and human decisions.
Another challenge was making the workflow understandable. Connecting checkout, orders, manifests, product references and packing inspections created more steps than we initially expected. Simplifying this experience remains important work.
Accomplishments that we're proud of
We built a working PayPal Sandbox integration that verifies payment capture details and connects the paid order to a shipment and packing manifest.
We also implemented manifest fingerprinting, packing-photo hashing, Gemini-based visual inspection and a separate merchant review decision.
One meaningful demonstration involved an inspection where the AI recognised the visible mug and quantity but could not reliably verify the exact catalogue design. The merchant chose Needs Correction, preventing an uncertain visual match from becoming an automatic approval.
We are proud that the prototype records the AI's observations separately from the merchant's decision rather than treating AI output as unquestionable.
What we learned
We learned that integrating AI into a business workflow is not simply about identifying objects in an image. The system must understand the expected order, preserve the reference information and communicate uncertainty clearly.
We also learned that payment verification, product identification and packing evidence need to be connected carefully if they are to be useful to merchants.
Most importantly, AI should assist a human decision rather than replace it when the available visual evidence is insufficient.
What's next for ParcelProof
Our next priority is to simplify and validate the core packing workflow with merchants who manually fulfil physical orders.
We want to complete and test the correction-and-reinspection journey, improve the speed and clarity of the packing experience, and explore barcode-assisted product identification alongside AI vision.
We also plan to investigate individual merchant accounts and stronger evidence handling.
The small-merchant customer profile is still a hypothesis, and we have not yet measured a reduction in packing errors, returns or support costs. Conversations with real merchants will help us determine whether this problem is frequent enough to justify the workflow and which features genuinely provide value.
Built With
- ai
- computer-vision
- docker
- fastapi
- google-gemini
- paypal
- postgresql
- python
- react
- supabase
- vite
