Teaching AI to Read a Letter of Credit

Checking commercial documents against a Letter of Credit — invoices, bills of lading, packing lists, certificates of origin — against the 14 core UCP 600 rules is detail-heavy work that traditionally depends on a small number of trained specialists reading dense SWIFT MT700 messages by hand. That made first-pass review slow, hard to scale during busy periods, and inconsistent when different reviewers applied the rules slightly differently.

InsightTech built Arkem: a system that parses the Letter of Credit directly from its SWIFT MT700 message, applies an automated 14-rule UCP 600 engine, and uses OCR plus an LLM to read uploaded commercial documents and cross-check them against the LC's terms — streaming findings back as the analysis runs rather than making the reviewer wait for a final report. Arkem's rule engine and MT700 parser proved reliable enough in production to become the foundation for TradeCheck, a multi-tenant SaaS rebuild that adds schema-per-tenant isolation and cloud LLM vision for document analysis.