FINTECH · SPONSOR BANKING · BaaS · FINANCIAL CRIME & COMPLIANCE
SEE THE SYSTEM.
BUILD THE ANSWER.
I have worked financial crime from inside a sponsor bank and from the fintech side of the same industry — partner risk at Stripe on assignment through AML RightSource, then BSA and partner oversight at a chartered BaaS bank. I know where the picture breaks between a bank, its middleware and its fintechs, because I have sat in all three conversations. So I built the platform that closes it.
SCROLL TO INITIATE ↓
THE ECOSYSTEM // WHERE I WORK
I DIDN'T STUDY THIS ECOSYSTEM.
I WORK INSIDE IT.
Banking-as-a-service, embedded finance and digital banking. Financial crime operations inside a sponsor bank carrying fintech program risk — and, before that, partner and third-party risk operations at platform scale on the Stripe engagement, on assignment through AML RightSource. The bank side and the fintech side, from inside both.
LAYER_01SPONSOR BANKThe charter. Where the risk finally lands.
LAYER_02MIDDLEWARELeast visible. Most often assumed to be someone else's.
LAYER_03FINTECH PROGRAMOwns the customer. Carries obligations it did not write.
LAYER_04CONSUMERThe end of the chain the industry treats as an externality.
THE PROBLEM
WHAT IS ACTUALLY BROKEN — AND WHY IT IS BREAKING NOW
WHY NOW // THE WINDOW
THE ADVERSARY GOT AI FIRST.
HE HAS NO COMPLIANCE DEPARTMENT.
Four things happened at once, and they did not happen to the same side evenly. Fraud industrialized. Money became instant. Oversight became something you must prove rather than intend. And the people attacking the system adopted new capability faster than the institutions defending it — because nothing slows them down.
Organized, funded and tooled. Adversaries share working playbooks faster than institutions share typologies, and they iterate without a change-control board.
Settlement collapsed from days to seconds. A decision window that once allowed review, escalation and a second opinion now allows one call.
Regulators moved from expecting intent to requiring evidence. Every control now has to be demonstrable, repeatable and defensible after the fact.
The governance that makes an institution trustworthy is the same governance that stops it moving at adversary speed. That is not a flaw. It is the constraint the answer has to respect.
THE WIDENING GAP
Schematic. Direction, not measurement.
THE RESOLUTION
CLOSING IT TAKES SOMEONE
WHO HAS DONE BOTH JOBS.
Someone who has worked the case, run the control, sat across from the examiner — and can build the system. For thirty years that person did not exist in one place, because the people who understood the problem could not build, and the people who could build had never worked a case.
That is the part that just changed. Doing this used to require an engineering team. It doesn't now.
INTELLIGENCE BRIEF // WHY SYSTEMS FAIL
THE PROBLEM ISN'T
A LACK OF TECHNOLOGY.
ASSESSMENT // ACTIVE
DATA≠INTELLIGENCE
AI≠EXPERTISE
TECH≠JUDGMENT
SME + DATA + TECH + AI→OPERATIONAL INTELLIGENCE
FAILURE_01 // FRAGMENTATION
THE DATA EXISTS.
THE PICTURE DOESN'T.
Alerts, customer history, documents, devices, partner communications, prior decisions and external intelligence may all exist — but not in the same operating context. Analysts reconstruct the picture by hand, one case at a time.
IMPACT duplicated work · inconsistent decisions · missed relationships
THE DIFFERENCE // SME → SYSTEM
KNOWLEDGE ISN'T A DATASET.
IT'S HOW YOU INTERPRET ONE.
The objective is not to pour decades of experience into a model. It is to encode the operating judgment behind the questions: what matters, what connects, what is missing, what should happen next, and when a human must decide.
HUMAN SME
EXPERIENCE
investigation · banking · fraud · AML · operations · BaaS
ENCODE OPERATING JUDGMENT
INTELLIGENCE LAYER
SPHERE
context · orchestration · memory · reasoning
DECISION THESIS // UNDERSTAND BEFORE EXIT
MORE DEBANKING ISN'T THE ANSWER.
BETTER UNDERSTANDING IS.
Some relationships absolutely need to end. But account closure should not become the default substitute for understanding the customer, the behavior, the history, the relationships and the actual source of risk.
Better understanding creates better decisions. Sometimes that decision is exit. Other times it may be a more precise intervention: restrict, monitor, escalate, remediate, educate or redesign the control around the risk.
Which means the appeal has to be real work, not a formality. A position — keep it open, or close it — is only worth as much as what stands behind it. Activity history that explains the pattern. EDD already on file. A clean complaint record. Source of funds. Licensing. The business model the numbers actually describe. That is what an appeal is made of, and it is what an examiner, a sponsor bank or a customer is entitled to see.
THIS IS A POINT OF VIEW ABOUT DECISION QUALITY — NOT A CLAIM THAT EVERY ACCOUNT SHOULD STAY OPEN.
BETTER DECISION MODEL
01UNDERSTANDcustomer · history · relationships · context
→
02DECIDEactual risk · proportional response · accountable judgment
→
03INTERVENEmonitor · restrict · escalate · remediate · close
WHAT AN APPEAL HAS TO CARRY
- THE FACTSactivity history, EDD on file, source of funds, licensing, complaint record
- THE ARGUMENTwhy the pattern is consistent with the business the customer actually runs
- THE ALTERNATIVEthe proportional control that manages the risk short of closing
- THE RECORDrationale and adverse-action evidence documented whichever way it lands
Appeal a decline the same way. A rejected application is a file, not a verdict — rebuilt on evidence, escalated where the facts support it.
BLUNT MODEL
ALERT → RESTRICT → CLOSE
BETTER MODEL
UNDERSTAND → DECIDE → INTERVENE
WHY ME
THE PERSON THIS PROBLEM SPENT THIRTY YEARS BUILDING
BUILDER DNA // PATTERN DETECTED
THE JOB TITLES CHANGED.
THE PATTERN DIDN'T.
I have spent most of my life taking systems apart, understanding what makes them work, and building a better version.
01 / INVESTIGATION
SEE WHAT OTHERS MISS.
Evidence, interviews, surveillance, behavior, patterns and consequence. Learn to follow what is real, not what is convenient.
02 / OPERATIONS
BUILD FROM ZERO.
Shipping, receiving, purchasing, inventory, subcontractors, teams, procedures, workflows and businesses. Structure turns effort into scale.
03 / FINANCIAL SYSTEMS
UNDERSTAND THE MONEY.
Banking, financial services, insurance, KYC, AML and regulated products created the financial context behind the investigations.
04 / FINTECH + BaaS
UNDERSTAND THE ECOSYSTEM.
Fintechs, sponsor banks, middleware, partner oversight and financial-crime operations reveal where intelligence breaks across organizational boundaries.
VENTURE HISTORY // SELECTED SIGNALS
BUILD. LEARN. EXIT. FAIL. BUILD AGAIN.
No résumé timeline. Just the experiences that changed how I build.
LEADERSHIP // WHAT I HAVE OWNED
FIVE TEAMS BUILT FROM ZERO.
NONE INHERITED.
Hiring, training, the standard they worked to, and the outcome. Different industries, same job: decide what good looks like, build the people and the process to produce it, and answer for the result.
Recruited and trained the team, wrote the investigative procedures and evidence standards, carried cases through to court testimony. Firm later sold.
Two teams hired and trained, plus subcontractors and a full site buildout from lease to opening day. Grew to $1.6M and Premier Store recognition.
Built and licensed the team, owned regulatory compliance and production standards for the agency. Profitable, and later sold.
Integrated compliance into onboarding, service and advisory work. Final decision-maker on escalated matters.
Selected to lead. Built the partner risk program end to end with Stripe risk leadership, then trained and transitioned it to their internal team. It went into production.
Not one of these was handed to me running. Every one started with an empty room and a standard that had to be written before anyone could be held to it.
BACKGROUND SIGNALS
UT AUSTIN COCKRELL // CIVIL ENGINEERING, ATTENDED
BUSINESS DEGREE // COMPLETED LATER
FORMER FINRA // 6 · 7 · 63 · 66
MULTI-STATE INSURANCE LICENSURE
CAMS // FINANCIAL CRIME
MILITARY + LAW ENFORCEMENT
SYSTEMS I BUILD // FROM CHAOS TO STRUCTURE
ARCHITECT THE MODEL.
BUILD IT. SCALE IT.
Sometimes the problem is a department. Sometimes it is a workflow, a reporting structure, a case process, a partner-oversight function or a technology layer. The common pattern is the same: understand the real problem, architect the operating model, build the missing pieces, then make the system repeatable and able to scale.
BUILD_01 // UNIT ARCHITECTURE
DESIGN THE UNIT.
NOT JUST THE TASKS.
A strong function needs more than analysts. It needs roles, escalation paths, ownership, handoffs, governance, reporting, metrics, quality controls and a clear reason for existing.
ROLES + RESPONSIBILITIES
ESCALATION MODEL
WORK QUEUES
QUALITY + GOVERNANCE
INTERACTIVE CONCEPT // UNIT ARCHITECTURE
LIVE BLUEPRINT
FUNCTION BLUEPRINT
INTAKEREVIEWESCALATEREPORT
OPERATING PRINCIPLES // HOW I APPROACH THE BUILD
THE TECHNOLOGY SHOULD
MAKE THE THINKING BETTER.
Do not automate a bad process simply because automation is available.
An alert, case, customer or dataset is usually only one piece of the operating system around it.
Rules, queries, known history and deterministic logic should solve what they can before AI is asked to reason.
A closed investigation should become future context, not forgotten history.
Machines should assemble, connect and accelerate. Consequential decisions still need accountable human authority.
FOUNDER STATEMENT // PERSONAL OPERATING SYSTEM
I HAVE NEVER BEEN VERY INTERESTED IN WORKING AROUND A BROKEN SYSTEM. I WANT TO UNDERSTAND WHY IT IS BROKEN — THEN BUILD THE BETTER ONE.
Gryphon Sentinel is not a departure from that history. It is what that history has been leading toward — not an AI idea looking for a financial-crime use case.
WHAT I BUILT
FROM THE THINKING TO THE WORKING SOFTWARE
CURRENT BUILD // THE PLATFORM
GRYPHON SENTINEL™
BUILT FOR THE FINTECH.
This is not a pitch. It is the clearest evidence I can give you of how I work.
Most financial-crime tooling is built for the bank. This is built for the company on the other end of the relationship — the fintech that owns the customer, carries obligations it did not write, and has to prove its program to a sponsor bank that can slow every launch. Working software, not a specification.
WHY I BUILT THIS ONE
I spent three years inside a sponsor bank watching fintech programs struggle with the same things: RFIs that stall a launch, evidence assembled from nothing before every exam, oversight questions arriving without context, and decisions made about their customers that they could not see or answer. I built it on my own time, from that vantage point — for the side of the relationship with the least visibility and the most to prove.
RFIs, escalations, 314(b) information sharing, refer-up packages and monthly MIS in one relationship view. The fintech refers up; the bank owns SAR decisioning. Coordination and advisory only.
The program surface the sponsor bank actually sees: posture, program, documents, monitoring, findings. Every control drills to its four layers — policy, procedure, control, evidence — with the evidence harvested out of daily operations rather than assembled before an exam.
The guided workspace for the bank relationship. A change request — adding a product, a rail, a card — triggers a fresh product-driven RFI package with due dates and open questions. AI guides and pre-checks. It never approves.
Continuous review across the fintech’s own book instead of periodic sweeps — pull builder, run queue, cadence and cycle activity, with scoring the fintech owns.
The front door and the second look. A decline is not the end of the file — rejections get appealed on facts, rebuilt with substantiated evidence, and the disputable ones become a reconsideration package to the sponsor bank.
Restrict, close, keep open, appeal, offboard. Every proposed closure is argued both ways before anyone acts, and the burden sits on the closure — not on the customer.
THE PART I CARE MOST ABOUT
AN ACCOUNT DOES NOT CLOSE BECAUSE A FLAG FIRED.
Closures and onboarding rejections get appealed, and they get appealed hard. Not with an opinion — with substantiated information: activity history, EDD already on file, complaint record, source of funds, licensing, the business model the numbers actually describe. Every proposed closure is argued on one screen in both directions, reasons to keep open against reasons to close, and the burden sits on the closure.
KEEP IT OPEN
Where the facts support the account, the system builds the case to keep it open — enhanced watch, a tighter review cadence, a real control — instead of defaulting to exit because exit is easier to defend.
OVERTURN THE REJECTION
A decline gets a second look on the merits. The file is rebuilt with evidence to find a defensible reason to accept, and the disputable ones become a reconsideration package to the sponsor bank.
DOCUMENT IT EITHER WAY
When the answer is still no, the rationale and the adverse-action evidence are on the record — defensible to an examiner, and answerable to the person whose account it was.
DE-RISKING BY REFLEX IS THE ONE DECISION NOBODY EVER HAS TO JUSTIFY. THIS MAKES THEM JUSTIFY IT.
BUILT BANK-AGNOSTIC
The sponsor bank is referenced generically throughout. The same program, evidence and coordination surface runs against a second bank — or two at once — without a rebuild. That was a design decision, not an afterthought.
PRIVACY WALL
The platform sees the fintech’s side only. Requests, due dates, questions, decisions and what comes next — never the bank’s internal notes or its risk ratings. Connecting the picture without pretending the boundary does not exist.
HOW IT GOT BUILT
Domain expertise defines the problem and the decision architecture. Models, data, APIs, automation and development tooling turn that architecture into working software. React/Vite front end, cloud deployed, no engineering team.
01DOMAIN EXPERTISEwhat actually matters
→
02ARCHITECTUREoperating model + decisions
→
03AI-ASSISTED BUILDcode + workflows + automation
→
04TEST + DEPLOYmake it real
CLAUDE CODECOWORKMCPCHATGPTHATZ AI
REACT / VITEGITHUBCLOUDFLAREAPIsSNOWFLAKE
POWER BIUNIT21UNITFOOTPRINTSARDINE
ALLOY
RECENT PROOF // AI-ENABLED FRAUD WORKFLOWS · MULTI-AGENT DEVELOPMENT · API-CONNECTED PRODUCT BUILD · PRIVATE GS DEMO
THE LAYER NOBODY BUILDS FOR
THE BANK, THE MIDDLEWARE AND THE FINTECH
ALL HAVE SOMEONE ARGUING FOR THEM.
Each of those layers owns different information, different duties and different permissions, and each one has a vendor, a counsel and a compliance function protecting its exposure. There is a fourth layer in every one of these diagrams, and it is the only one with nobody in the room.
WHY I KEEP COMING BACK TO THIS
Before any of this, I spent years on the retail side of the business — a licensed business bank officer at JPMorgan Chase, then consumer advisory at an insurance agency I built and owned. Sitting across a desk from a real person and telling them what an institution had decided about their money.
Every layer above the consumer optimizes for its own exposure. The person the money actually left gets no seat in that conversation, no visibility into the decision, and no way to correct the record.
Every decision on the screens above eventually lands on that person. That is why the appeal matters, and it is why I build the way I do.
COLLABORATION // OPEN
GIVE ME THE HARD PROBLEM.
I am at my best where the answer is not already sitting in a procedure manual — where the environment has to be understood, redesigned and built. If you run a fintech, a sponsor bank or a middleware platform and the compliance picture is fragmented, that is the conversation I want.
START A CONVERSATION DOWNLOAD BUILDER PROFILE LINKEDIN
THE SHAPE IS NEGOTIABLE
LEAD THE FUNCTIONBUILD THE PROGRAMDESIGN PARTNERDEPLOY THE PLATFORMINVEST
Founder, operator, design partner, strategic customer, institutional build, advisor, investor — the label matters less than whether the problem, the people and the incentives line up.
FIELD NOTES // THINKING IN PUBLIC
IDEAS WORTH
ARGUING ABOUT.
AI IS NOT INTELLIGENCE.
Why adding a model to a broken operating system does not create expertise.
THE INSTITUTIONAL MEMORY PROBLEM.
Financial institutions know more than their systems can remember.
THE BaaS INTELLIGENCE GAP.
Bank, middleware and fintech each see a different piece of the same risk.
BUILDERS NEED DOMAIN EXPERTS.
The dataset is only the beginning. The hard part is knowing what to do with it.