AML red flag checking in trade finance was slow, manual, and error-prone
Anti-money laundering (AML) compliance in trade finance is one of the most document-intensive processes in banking. Compliance teams must manually review trade documents โ letters of credit, invoices, bills of lading โ to identify red flags: price manipulation, phantom shipments, duplicate invoices, sanctioned counterparties, and inconsistent data across documents.
Why this matters: Trade-based money laundering (TBML) is estimated to be one of the largest methods used to move illicit funds globally. Yet most banks still rely on rule-based systems and manual analyst review โ slow, costly, and unable to scale with document volumes. A Gen AI solution that could read, cross-reference, and flag anomalies across multiple trade documents in seconds would be transformative.
BNP Paribas's APAC CIB Operations team saw this as a prime use case for Generative AI โ and the hackathon was the proving ground to test whether it could actually work.
๐ธ Hackathon presentation โ BNP Paribas, Hong Kong & Singapore
[ Add your presentation photos here ]
Team lead โ from problem framing to live demo
I led a cross-functional team of 4โ6 people through the full hackathon cycle: identifying the use case, defining the solution architecture, coordinating build across team members, and presenting to the judging panel. This was a GTM and innovation leadership challenge as much as a technical one โ the winning teams weren't just those who built something clever, but those who clearly articulated the business case and demonstrated real impact.
The hardest part wasn't the AI โ it was framing the problem so precisely that the solution could be evaluated in a live demo under time pressure.
A Gen AI tool that reads trade documents and flags AML risks automatically
Mapped the existing AML review workflow for trade finance documents โ identifying where analysts spent the most time and where errors were most likely. Chose the highest-impact entry point: cross-document red flag detection.
Designed a Gen AI pipeline that ingests trade documents (letters of credit, invoices, shipping records), extracts structured data, and applies a library of AML red flag rules โ checking for price anomalies, counterparty mismatches, and document inconsistencies.
Led the team in building the prototype, with a strong focus on prompt engineering to ensure the LLM produced structured, auditable outputs โ not just summaries, but flagged findings with rationale that a compliance officer could act on.
Presented to senior BNP Paribas judges across Hong Kong and Singapore. Demonstrated the tool on real trade finance document scenarios, showing time savings and flag accuracy versus the manual baseline. Won both regional awards.
Dual regional champion โ and a pathway to production
What winning taught me about AI GTM in financial services
The judges didn't just evaluate the technology โ they evaluated whether it could survive contact with compliance, operations, and IT. The teams that lost often built impressive demos that no one could explain how to deploy. We won because we designed for the workflow first, the AI second.
This experience shaped how I approach AI GTM in banking: the bottleneck is never the model. It's the change management, the audit trail, the explainability to regulators, and the integration into existing ops processes. A Gen AI solution in financial services needs a compliance narrative as much as it needs a technical one.
Post-hackathon, being named AI Catalyst gave me a mandate to apply this thinking at scale โ identifying Gen AI use cases across APAC CIB and shepherding them from prototype to production. The hackathon was the pitch; the catalyst role was the GTM.