Consumer provenance chat
A shopper photographs a product's back label and asks where it came from. The answer comes from real supply-chain records, not the model's imagination.
LLMOCRNode.jsMongoDB
The idea
Shoppers want to know where a product came from. The traceability data already exists, so the chat’s job is to find the right records and explain them in plain language without inventing anything.
How it works
- The shopper takes a photo of the back label.
- OCR pulls the identifying details off the label.
- The backend resolves the supply-chain batch those details belong to.
- An LLM narrates the answer, grounded only in that batch’s records.
Decisions
- Failover across 4 models. If one provider is slow or down, the next one answers. The chat never depends on a single vendor.
- Context on the client. After the first answer, the conversation context lives in the browser. A follow-up question costs 1 LLM call and 0 backend calls.