Commodity trade finance, 8 years of history, regulated data
The group financed commodity trades on a rolling cycle with its banks and ran the operation out of spreadsheets. Simple questions, like what do we owe today, took days of digging. Confidential records were legally bound to stay in-country, so a public cloud AI was off the table.
We delivered a live analytics and reporting layer across the whole business, then added an AI layer that reasons over the data on-premises. A locally hosted large language model scores its confidence green, amber or red and learns from every correction.
8 years of manual reconciliation, roughly 1,600 transactions across about 100 monthly snapshots, now runs as a live system.
Measured
Outstanding bank position and gross margin across the full book are visible instantly.
Measured
The on-premises engine is targeted to handle 85 to 95% of the judgment work automatically.
Directional
Data-residency compliance by design: nothing ever leaves the jurisdiction.
Measured
A web app that parses the full transaction history into analytics with automated data-quality checks, plus a locally hosted open-weight LLM served in-region with embeddings, confidence scoring, human review and adaptive learning.
Next step
Book a discovery call. First quote in 3 hours.
