CargoLens

About CargoLens Inspiration Cross-border orders are often reviewed across separate spreadsheets, documents, watchlists, and customs rules. A missing importer detail, an unclear product description, or an inconsistent HS code may be discovered only after several people have handled the order. We wanted to help teams identify these issues earlier and show "why" an order needs attention. What we built CargoLens is a prototype for preliminary compliance review of orders shipped from China to the United States. Its workflow is designed around three checks: trading parties, invoice and product details, and importer qualifications. The working prototype checks structured orders for missing information, arithmetic inconsistencies, HS code candidates, description conflicts, and importer details. We also prepared Coze large-language-model nodes to assess whether a product description is consistent with its declared HS code and to flag risks that fixed rules may miss. Results follow a common JSON format with findings and evidence, so uncertain cases can be sent to a human reviewer.

How we built it We defined a shared order schema, implemented deterministic checks in Python, and arranged them as modules for a Coze workflow. For example, the invoice module verifies:

     A separate aggregation step validates model output and preserves any issues found by the rules. We tested the rule components with synthetic orders covering normal, incomplete, inconsistent, and potentially risky cases.

Challenges and lessons learned The hardest design choice was deciding what the system could responsibly conclude. An HS match is only a candidate check, and an unusual price is a reason to review evidence—not proof of a false declaration. Reliable comparison data and external verification services were also limited, so our demonstration price baseline is explicitly marked as "mock". We learned that language models are useful for spotting semantic inconsistencies, while arithmetic, required fields, and failure handling need explicit rules. CargoLens is built to support a reviewer’s decision, with the final compliance judgment remaining with qualified people.