4–6mo→3wk
AI approval cycle, once governance is in place
Why teams bring me in
I have worked in every block I now build for.
Engineering, lending, cards, payments, wealth and financial crime, then the bank itself as its first AI hire. That is why a CTO, a head of credit or a risk committee will sit down with me: I understand how their organisation makes decisions, and where it gets stuck.
Core engineering
HSBC · GE Money
Lending systems and self-service platforms from the inside.
Cards & credit
Discover
Card products, transaction processing and settlement at scale.
Wealth & risk
RBC
Regulated wealth product across multiple jurisdictions.
Lending
Wonga
Credit, scoring and decision engines through FCA authorisation.
Payments
Thredd
£4.2B card-issuing platform, real-time financial crime, 100+ fintechs. Led a 15-person product org across London, Singapore and Sydney.
Financial crime
ComplyStream
AI compliance platform, co-founded and shipped to FCA customers.
The bank
UK specialist bank
First AI hire. Two systems in production, governed and signed off.
The build layer
regstack
Open-source compliance agents. The whole journey, codified.
Track record
Engineer. Operator. Founder.
Now
Vyaya
Founder · AI build & consulting
Shipping production AI and the governance to clear it, for regulated finance. Maintainer of regstack. Trading as Samantaray Digital Ltd.
2025
UK specialist bank
AI Strike Lead · dual FCA/PRA
First AI build lead on the strike team. 19 use cases across 5 BUs, 2 in production, one Trust Layer that cut approvals from six months to three weeks.
2024
ComplyStream
Co-founder & CPO
AI financial-crime platform, zero to FCA-regulated customers. £850K pre-seed, Google for Startups AI, NVIDIA Inception. 500 to 25 alerts a day, 85% fewer false positives.
2018
Thredd (ex-GPS)
VP Product · P&L owner
£4.2B card platform, 60 to 100+ fintechs, 99.99% uptime. Built NLP developer tooling before GPT was public.
2016
Wonga
Head of Lending Product UK
1.2M applications a year, decided in 90 seconds. Rebuilt the credit models through an 18-month FCA authorisation, serving 2M+ throughout.
2012
RBC Wealth Management
Product tech lead · Toronto
Regulated wealth product in a global bank. Where the second-line instinct came from.
2008
Discover
Cards & credit · Chicago
Card products and credit systems at scale in US consumer finance.
2004
HSBC
Software engineer to tech lead · Chicago
Where it started. Lending engineering, the foundation under all of it.
What it looks like in practice
Shipped with the team. Measured honestly.
A few engagements where AI went past the demo and into something a regulated business now runs. The numbers are what the desks actually saw, not a launch headline.
TinyFish Accelerator · 2026
Compliance OS, built solo, final 6 of 2,000+
An agentic KYC and compliance build that turns manual onboarding and review into audit-ready, regulator-defensible decisions. Selected as a Demo Day finalist and pitched live to a leading SF VC and the TinyFish leadership team.
6 / 2,000+
Demo Day finalists, from applicants
solo
built and pitched in weeks
UK specialist bank · first AI hire
From 19 ideas to two in production
Embedded with business, second-line and IT to find which use cases were real. Built two into production, complaints drafting and a valuation assistant, and authored the Trust Layer that got them signed off. It became the bank's standard for every AI deployment after.
4–6mo → 3wk
approval cycle, once governed
2 / 19
shipped, not boiled the ocean
SME lender · credit workflow
The work around the score
Their ML does the scoring. The slow part is everything around it: extraction, the credit paper, the audit trail, monitoring. Scoped that into phased agentic use cases the credit team could trust, with the decisioning kept sovereign and governance sized for their warehouse lender.
prove → prod
phased, with their engineers
sovereign
decisioning stays theirs
ComplyStream · co-founder & CPO
A compliance platform, zero to FCA customers
£850K pre-seed from Cornerstone, Ascension and operators at Monzo and ClearBank. Google for Startups AI, NVIDIA Inception. Built the product and shipped it to FCA-regulated customers as co-founder and CPO, before moving on to build independently.
500 → 25
alerts a day, the right ones
CPO
co-founder & product lead
Payments & advisory · neobank build
Named on the build, not just the deck
Brought in as the regulated-AI and payments voice on an AI-native neobank build: BaaS selection, card programme, AI architecture, hypercare. The reason teams ask is rare coverage of payments, AI and lending in one head, with £4.2B of card processing behind it.
16 wk
embedded across the MVP
end to end
selection to hypercare
Open source
I ship in the open, so the proof is public.
Client work stays private, so I build open-source compliance tooling in my own time. It is where you can actually read my code and see how I think about regulated AI. regstack is the home for it.
kyc-analyst ↗live
Full KYC onboarding with independent verification, deterministic four-factor risk scoring, immutable audit trail, 17-section report. 33+ stars, organic forks from IPO'd fintechs.
★ 33+v1.1 · public
risk-readiness-checker ↗v2
Three-pillar engine: regulatory mapping, policy framework, data orchestration, with a CRO / MLRO / FCA expert-challenge layer.
14 stagegatesdashboard
WhtrBT ↗built
Trust infrastructure for AI-generated code. 95 patterns mapping EU AI Act Article 13 and PCI-DSS into developer-native compliance.
95 patternsAI Act + PCI
fraud-mcp ↗built
A 10-tool FastMCP server wired to a live payments API for real-time fraud signals inside an agent loop.
10 toolsFastMCP
Recognised for this work
Invited where regulated AI is being figured out.
I get asked onto panels and into rooms on agentic payments and regulated builds because I have actually shipped them. I also write about it, in the open.
↳ Speaking & recognition
Agentic payments & regulated builds, panels and talks at Fintech Guild, Zero to Agent (Vercel HQ) and CCCL.
Anthropic Builder Summit London, inaugural cohort.
TinyFish Accelerator, final 6 of 2,000+, pitched at Demo Day.
Google Cloud Blog feature, press in BusinessCloud.
↳ Not Another Fintech Alert
AI COGS repricing, what agent unit economics do to a compliance P&L. ↗ Rathbones / FCA s166, reading a skilled-person review as a product spec. ↗ The compliance-startup bear case, where funding meets a thin moat. ↗ Lovable data study, a practitioner's critique of the numbers. ↗Research
I publish the analysis, not just the take.
Interactive, data-led reports I research, build and deploy myself. The kind of primary-source work I bring to a build or a board.
Report · platform analysisLive ↗
The Distribution OS
Claude's plugin registry as a distribution layer: who gets anchored, amplified or absorbed across 9,000+ MCP servers. Nine findings on software, talent, commercials and regulation.
Deployed on Netlify · interactive
Report · community signalLive ↗
What AI Users Actually Think
Twelve months of primary signal from Reddit, Hacker News, GitHub and Stack Overflow. Agent pilot failure rates, the open-source model shift, MCP security, context engineering.
Deployed on Netlify · interactive
The build
One architecture. Every regulated use case.
How I take a workflow problem to production a regulator, a board or a buyer will stand behind, governed from day one.
Who I build for
Banks & fintechsRegTech scale-upsPE & growth fundsConsultancies & SIs
Regulated finance · governed from day one
Embedded with your engineers, data and risk people
Sources
Documents, transactions, customer data, market & risk, policy
core-banking & card rails · MS Document Intelligence · PostgreSQL / TimescaleDB
Ingest & retrieve
Connectors / MCP, pipelines, RAG + vector store
FastMCP & MCP servers · FastAPI · RAG · Redis
Reason & build
Agent orchestration, multi-model (Bedrock / Vertex / Azure), evals
Claude (Opus/Sonnet) · Gemini · Bedrock / Vertex / Azure OpenAI · sub-agents, skills, hooks · custom ML · evals
In production · across financial services
KYC / AML & fin-crimeCredit & underwritingPayments & fraudServicing, claims & opsCompliance & reportingRisk & model governance
Shipped on: Docker · GitHub Actions · Cloud Run · React / Next.js dashboards
+ collections, treasury, wealth & advice, insurance underwriting, AML transaction monitoring, reg-change management (placeholder, tell me your priorities)
Trust Layer, built on regstack (open-source): governance, guardrails, evals, audit log, human-in-the-loop, model-risk, lineage
Signed off: a regulator, a board or a buyer stands behind it
How I engage: Discovery sprint → Build & ship → Fractional AI / CPO → Named on your bid. Plus advisory & board for fintech founders and early-stage AI teams.
What I build with
I came up as an engineer and never left the codebase. I build multi-cloud, multi-model systems hands-on: I have shipped custom Python ML and LLM orchestration across AWS Bedrock, Google Vertex and Azure OpenAI, and I build agents in the modern Claude and Gemini stack. I bring in specialists where it helps, but I am not a slideware operator.
Languages
Python (FastAPI, FastMCP, ML) · TypeScript · JavaScript / Node · SQL · Bash
LLMs & models
Claude (Sonnet, Opus) · Gemini (3 Pro, 2.5 Flash, AI Studio) · Llama via Bedrock · custom ML (scikit-learn, transformers, entity extraction) · RAG · multi-model orchestration · evals
Cloud & AI platforms
AWS Bedrock · Google Vertex AI & Cloud Run (GCP) · Azure OpenAI, Foundry & MS Document Intelligence · genuinely multi-cloud, not single-vendor
Agents & AI-native tooling
Claude Code · Cowork plugins · MCP & FastMCP servers · sub-agents, skills, slash commands, hooks · @google/genai SDK · context engineering
Data & infrastructure
PostgreSQL · TimescaleDB · SQLite · Redis · Docker · GitHub Actions · Git · Vercel · Netlify · Railway
Build, design & research
React & Next.js · Tailwind · Recharts & D3 · Mermaid · Excalidraw · Figma · Obsidian · REST & API-first design
How I work
With your team, from discovery to production.
I do not parachute in with an agent and leave. I embed with the engineers, data and risk people who own the system, and move in phases so value and confidence build together.
01 · Prove
Find the real workflow
Map where the work actually gets stuck, not where a demo looks good. Pick one use case, build a working prototype, agree how we measure it.
Weeks · low cost · clear go / no-go
02 · Pilot
Build it with the team
Ship into a real workflow with your engineers and data team. Wire the integrations, write the evals, stand up the audit trail, governance designed in, not bolted on.
Measured on real desks · reviewed as we go
03 · Production
Earn the sign-off
Trust scoring, escalation, logging and explainability the second line and Risk Committee will actually approve. Then hand it to an owner and scale to the next use case.
Regulator-grade · staggered ROI · yours to run
Where we start: most engagements open with a short paid discovery, usually a couple of weeks, to find the use case worth building and agree how we measure it. No long retainer before there is something working to point at.
The deal
What you get, and what you won't.
01
I ship code with your team, not a slide deck that recommends one.
02
An honest efficiency view, phased. No single hero number on the cover.
03
I tell you what will not work before you pay for it, not after.
04
I'll say no to a use case that can't clear review. That protects you.
Banks · fintechs · PSPs
Build a real agent and the governance to ship it, with your engineers.
RegTech scale-ups
Fractional product and AI leadership, founder to operator, through to exit.
Consultancies & SIs
The named regulated-AI specialist on your bid, with shipped proof behind it.
PE & growth funds
AI value creation across portfolio companies, from thesis to production.
Ways to work together
Four ways in. Each one named, scoped and shippable.
No open-ended retainer before there is something working to point at. Most teams start with a fixed-fee discovery, then scale into a build once the use case has earned it.
Discovery Sprint
Fixed fee · 2–3 weeksFind the one use case worth building, prove it, and agree how we measure it, before any long commitment.
- A scored use-case shortlist across your business
- One working prototype on a real workflow
- A measurement plan and a clear go / no-go
Build & Ship
Embedded · 12–16 weeksTake that use case into production alongside your engineers, data and risk people, governed from day one, not bolted on.
- One use case live on a real desk
- The Trust Layer: trust scoring, escalation, logging, audit trail
- A sign-off pack for second line and Risk Committee, plus handover
Fractional AI / CPO
Monthly retainerSenior AI and product leadership on tap, for banks, fintechs, RegTech scale-ups, consultancies and PE-backed teams taking AI from prototype to production.
- Strategy, roadmap and hands-on build, not just advice
- Governance and a board-ready narrative for every release
→ Two to three at a time.
Named on your bid
Per engagementFor consultancies, SIs and PE funds: I am the named regulated-AI specialist on the enterprise pitch or the diligence, with shipped proof behind the name.
- A credible regulated-AI voice on the bid or the deal
- Portfolio AI value creation, thesis to production
- Operator inside a PE-backed scale-up, with M&A diligence alongside Bain and Oliver Wyman.
On pricing: discovery is a fixed fee; builds and retainers are scoped to the work. Indicative numbers on the intro call.