AI is not a tool. It's a way of thinking about problems.
I started exploring AI in 2019 โ not because it was trendy, but because I saw it as the most powerful lens for re-examining how work gets done. In banking operations, the default answer to complexity is more people, more process, more manual checks. My question was always: what if we didn't have to do it that way?
The most valuable thing Gen AI gave me wasn't productivity. It was permission to ask: why are we still doing this by hand?
That mindset โ look at any repetitive, error-prone, human-intensive process and ask whether intelligence could be applied to it โ became the foundation of everything I built at BNP Paribas. Not as a technologist, but as someone who understood the business deeply enough to know where AI would actually matter.
AML compliance in trade finance โ a manual process crying out for AI
Anti-money laundering checks in trade finance are among the most document-intensive processes in banking. Compliance teams manually review stacks of trade documents โ letters of credit, invoices, bills of lading โ searching for red flags: price inconsistencies, phantom shipments, duplicate invoices, mismatched counterparty data. Each transaction could involve dozens of documents, reviewed by hand, under time pressure.
Why this was the right problem to pick: Trade-based money laundering is one of the least-detected forms of financial crime globally โ not because the signals aren't there, but because the volume of documents overwhelms the capacity for human review. A Gen AI system that could read, cross-reference, and surface anomalies across multiple documents in seconds wasn't just more efficient. It was a fundamentally different capability.
When BNP Paribas launched its internal global AI hackathon, I saw this as the moment to prove what I'd been building toward since 2019 โ that Gen AI, applied with the right problem framing and domain knowledge, could do something genuinely new in banking compliance.
Building fluency before it was expected
I didn't wait for a mandate. I started learning, building, and applying AI to real problems from 2019 โ well before most banking teams had a formal AI strategy, and years before Gen AI became mainstream.
As a fintech solution provider, designed a new product for banking partners: an NLP-based system for cross-referencing underlying trade documents. Real product design, real banking context โ four years before the Gen AI wave.
Started building with GPT-4 and open-source models. Developed prompt engineering skills and mapped real banking operations problems to Gen AI capabilities โ AML document review, trade finance ops, and compliance workflows.
Led a team of 4 to build a Gen AI pipeline that ingests trade finance documents, extracts structured data, and applies AML red flag logic โ returning flagged findings with rationale a compliance officer could act on. Won the APAC regional award against 500+ global participants.
Appointed by BNP Paribas to lead Gen AI rollout and deployment across APAC CIB Ops. Built the AI adoption framework that grew Gen AI usage from 0 to 60% across twelve countries in the region.
Continuously identifying new AI use cases, running pilots, and working with business leads to shift from "AI as experiment" to "AI as default approach." Currently exploring Agentic AI pilots in banking operations workflows.
Designing for the workflow, not just the technology
The winning insight wasn't technical โ it was operational. Most hackathon teams built impressive demos that no one could explain how to deploy in a regulated environment. We designed ours backwards: starting from what a compliance officer actually needed to see, what an auditor would require for traceability, and what IT would need to integrate it into existing systems.
Mapped the AML review workflow step by step. Identified that the bottleneck wasn't knowledge โ compliance officers knew what red flags looked like. The bottleneck was the time it took to manually cross-reference data across 10โ30 documents per transaction.
Designed a document ingestion and extraction pipeline using LLMs. The model reads trade documents, extracts structured fields (counterparties, amounts, goods descriptions, shipping routes), and cross-references them against a library of AML red flag patterns.
The hardest part was making the output usable by compliance teams. Generic LLM summaries weren't enough โ we needed flagged findings with specific citations, confidence levels, and reasoning. Spent significant time on prompt architecture to produce structured, auditable outputs.
Presented to senior judges by leading with the regulatory and operational case, not the technology. Demonstrated how the tool fit into existing compliance workflows, reduced review time, and produced outputs that satisfied audit requirements. Won both regional titles.
From hackathon win to organisation-wide transformation
AI transformation is a people problem, not a technology problem
The hardest part of driving AI adoption in a bank wasn't building the tools โ it was shifting the mindset of people who had done things the same way for decades. Compliance officers who had built their careers on manual expertise needed to see AI as an amplifier, not a replacement. Risk managers needed assurance that the outputs were auditable. Senior leadership needed a business case that went beyond efficiency.
What I learned is that successful AI transformation requires three things working simultaneously: a technically sound solution, deep domain knowledge to frame the right problems, and the human skills to bring people on the journey. I can do all three โ and that's the combination that's rare.
Starting in 2019 โ as a fintech solution provider designing NLP-based trade document comparison tools for banking partners โ gave me something most people don't have: genuine AI fluency earned through building real products, not just reading about them. I understand how these models behave, where they fail, how to prompt for reliable outputs, and how to design human-AI workflows that hold up in regulated environments. That's not a credential โ it's experience.