I build AI agents that take manual review off compliance and operations teams, and I get them past the desks that stop most of this work. Compliance, security, architecture, platform engineering. The agent runs in your cloud, on your architecture standards, cleared before anyone is asked to use it. Fourteen years inside banks and payment companies, on the other side of that table.
Different mandates, same problem. People reading files by hand because nothing else could be trusted to.
Scaling fintechs whose compliance function was built for a tenth of today's volume, and who need to clear a bank partner's due diligence.
Retail and commercial banks whose KYC, screening, and correspondent due diligence still run on analysts, email, and spreadsheets.
Money transmitters and payment firms carrying heavy compliance load per transaction. Correspondent onboarding, screening, and the manual review that scales badly as volume grows.
Supervisors and central banks weighing digital ID, reusable KYC, and interoperable fast payments. The infrastructure decisions that determine who gets into the financial system at all.
Compliance and operations headcount scales with volume. Agents do not. Institutions that get AI into production first will run the same book with a fraction of the manual review, and the ones that stall will stall on governance, not on models. That is the gap Pontarch closes.
Capability stopped being the constraint. Getting an agent inside a regulated institution's cloud, controls, and approval chain is the constraint now.
BSA/AML obligations, OFAC screening, and FATF correspondent due diligence expectations keep expanding. Headcount cannot keep pace with transaction growth.
Model risk guidance expects a decision to be explainable and reviewable. An agent that cannot show why it cleared or escalated a file will not survive an examiner, however good it is.
The demo lands. Then security, architecture, and platform engineering each ask a question nobody prepared for. That is where most AI work stops.
Pontarch Advisory is led by Rudy Rocourt. He architects the future of payments for unbanked populations around the world.
Rudy spent fourteen years in international finance and financial technology before founding the firm. The work took him through Citibank, BNY Mellon, and Viamericas, across the U.S., India, and the Caribbean. After the 2010 earthquake he returned home to head Citibank Haiti’s Treasury department, where he saw how much of the economy runs on diaspora remittances. He went on to found and run Jetli, a cross-border payments platform that delivered goods to families in Haiti within 48 hours with no transfer fees, backed by USAID’s Development Innovation Ventures.
That path shapes how he works now. Rudy has worked inside banks and he has shipped a payments product with real users and real regulators. He builds for the people who have to live with the system after the consultant leaves. He maps how a team actually works today before designing anything, and builds shoulder to shoulder with client engineers, to their architecture standards, so the product arrives already approved. He works as a forward deployed engineer. Embedded with your team, building in your cloud, to your standards.
One person is a fair thing to worry about. So the work is built so it does not depend on me. The code lives in your repositories, deploys through your pipelines, and runs in your cloud. Your engineers are on it from the first commit, not briefed at the end. If I disappeared in month four, your team would still have a system they helped build and can maintain. That is the design, not a reassurance.
A short list on purpose. These are the areas I have built in, not a catalogue.
Compliance analysts spend their days reading files, screening names against OFAC and sanctions lists, researching adverse media, and chasing missing documents. An agent can take most of that. The hard part is doing it so the audit trail survives an examination, so a reviewer can see why the agent cleared or escalated a file, and so false positives land somewhere a human can work them.
Onboarding a correspondent bank means collecting documents, shareholder and officer data, and screening every name attached. Most institutions run it over email. A self-service portal moves the work to the counterparty, validates as it goes, and exports files the compliance engine can actually consume.
Cross-border money movement is where the friction and the margin live. I have run treasury inside a bank, built a remittance product, and worked at a money transmitter. Corridor strategy, partner and correspondent networks, wallet and direct-to-bank payout, priced and sequenced against what the institution can actually deliver.
Most AI pilots die between the demo and production. Not because the model is wrong, but because the system never cleared security, never fit the architecture, and never got signed off by the people who have to live with it. This is the part nobody sells and everybody needs.
A prototype can be standing in a week. Getting it into production inside a regulated institution is the actual job.
Before anything is designed, I sit with the team and watch how a file actually moves. What gets checked, what gets skipped, where the judgment lives. The agent gets designed around that reality, not around a process diagram.
Your cloud, your backend, your existing tools, your security and governance review. I read your engineering standards and build to them from the first commit, so nothing has to be rebuilt to pass review later.
Compliance, security, architecture, and platform engineering each hold a veto. I get in front of all of them early. Teams do not reject new systems because they fear them. They reject what never cleared the room, and one holdout is enough.
I build shoulder to shoulder with your team so the system holds up after I leave. The goal is not a dependency on me. It is a capability your institution owns.
Pontarch is an advisory firm with a studio inside. The studio is where the build gets done. Nothing gets recommended that has not been built and tested first.
Client names withheld. Nothing here is a projection.
I name the model, not a price tag — every scope is different. I quote after a first conversation.
Book a 30-minute call. Bring one workflow that eats your team's hours and we will work out whether an agent can take it.
Book a 30-minute callOr email rudy.rocourt@pontarchadvisory.com