I redesign O2C, decisioning and control architectures across borders, so automation improves cash, control and operating leverage — not just throughput.
25+ years architecting Order-to-Cash and Finance operating models across European and Asian delivery environments: a $30M+ transformation portfolio, €3M+ in revenue-recovery solution architecture, 25+ enterprise go-lives, 500+ SLAs governed for 30+ global clients, and 140+ continuous-improvement initiatives. Now extended into local LLMs, agentic workflows and enterprise AI control architecture.
Finance transformation fails when workflow speed is mistaken for process integrity. The architecture has to preserve decision quality, controls, explainability and economic intent before AI is allowed to accelerate it — and it has to hold across entities, currencies, tax regimes and jurisdictions, not only inside the entity where it was designed.
The work sits between Finance, operations, product and technology: revenue logic, ERP orchestration, control design, AI decisioning and measurable operating outcomes. Automating a broken O2C cycle does not recover cash — it industrialises the leak and removes the manual checks that were quietly containing it.
I engineer Finance and technology environments that can grow across markets without sacrificing control, cash, decision quality or execution discipline.
Across 25+ years in O2C, AR and enterprise transformation, I have engineered platform implementations, operating-model redesigns and AI-enabled workflows across FMCG, Banking, Telecom, Pharma and Consumer Goods, in European and Asian delivery environments.
My value is structural: I identify where an operating model will fail under scale, then redesign and fortify the process, data, controls, technology and decision architecture so the failure mode is removed rather than repeatedly absorbed by the business.
I synthesize deep O2C/AR expertise with SaaS product delivery, enterprise platform architecture and hands-on AI engineering. I leverage local LLM inference, context-aware DLP, explainable credit-risk tooling and GenAI workflow design where they strengthen the operating model.
Evidence at scale: $30M+ transformation portfolio, €3M+ in revenue-recovery solutions architected around leakage and process failure, 25+ EU AR platform go-lives, 140+ continuous-improvement initiatives, $100K+ realised savings, and SLA/KPI architecture spanning 500+ SLAs for 30+ global clients.
Working systems that test how enterprise Finance and AI can retain control, explainability and ownership at the point where decisions are made.
A self-hosted, API-first credit-risk and AR platform with invoice-level data, behavioural scoring, automated dunning, payment prediction, 30/60/90-day cash-flow forecasting and a locally running AI assistant.
Credit decisions influence cash, customer treatment and working capital. They cannot become opaque simply because AI entered the workflow. RSK keeps the decision logic reproducible, grounds AI in live finance data and places inference inside an ERP-ready architecture rather than behind an uncontrolled cloud prompt.
S.T.O.P. Sentinel is a local-first clipboard DLP agent that intercepts clipboard mutations in real time, redacts credentials, API tokens, payment instruments and PII before paste, and maintains an encrypted audit trail. Native OS interception in C/Objective-C is combined with a Python policy engine using regex, Luhn validation and Shannon entropy heuristics.
Enterprise AI creates a new disclosure boundary at the point of paste. S.T.O.P. Sentinel moves control to that boundary, preventing sensitive data from leaving the governed environment instead of detecting exposure after it occurs — the difference between a control and an incident report.
An end-to-end local LLM training pipeline spanning data curation, tokenisation, transformer training, alignment and inference deployment — designed for custom GPT-style models without mandatory cloud dependency.
Enterprise AI strategy becomes structurally dependent when training data, inference economics and model access all sit outside the organisation's control. A local model path matters where data residency, specialised domain knowledge, economics or explainability justify ownership of the model lifecycle.
Applied research across finance integrity, AI control and high-efficiency data infrastructure — turning operating problems into architectures that can be tested, challenged and delivered at enterprise scale.
Impact statementConverts an ambiguous business problem into an explainable AI-or-not-AI decision, EU AI Act risk screen and actionable delivery plan before significant investment begins.
PMAID combines deterministic, editable decision rules with optional local llama.cpp assistance. It guides product managers through structured discovery, explains recommendations in plain English and generates eight delivery phases with tasks, gates, roles, risks, question-back prompts, timelines, risk heatmaps, RACI ownership and sponsor-ready exports. Project data remains local and no external AI API is required.
Impact statementArgues that sensitive-data disclosure to AI systems is an enforcement-point problem rather than a detection problem, and demonstrates prevention at the point of paste by combining native clipboard interception with a local policy engine, validated against an 83-rule synthetic evaluation corpus.
Intercepts clipboard mutations in real time, redacts credentials, API tokens, payment instruments and PII before paste, and maintains an encrypted audit trail using native OS interception with a Python policy engine built on regex, Luhn validation and Shannon entropy heuristics. Relevant where AI adoption outpaces the control model: data residency, endpoint governance and audit evidence across regulated, multi-jurisdiction environments.
Impact statementFrames AR fraud as an identity-layer failure rather than a detection shortfall, and proposes a blockchain-anchored decentralised identity architecture designed for implementation inside enterprise O2C programmes rather than as a parallel control stack.
Grounded in hands-on O2C delivery across 25+ European client go-lives, where counterparty identity fails first at the cross-border payment, remittance and bank-detail-change boundary — the point at which detection models are structurally too late.
Impact statementExamines whether model-assisted compression can reduce the infrastructure burden of large-scale image archives while preserving reconstruction fidelity and operational throughput.
Archive storage is a recurring, compounding line item in enterprise infrastructure cost. The architectural question is where model-assisted reconstruction is economically justified against retrieval latency and fidelity requirements.
Impact statementExplores a hybrid spatial-temporal codec architecture designed to balance compression efficiency, reconstruction fidelity and throughput for high-scale enterprise media environments.
Built end-to-end to test the design trade-off directly rather than argue it on paper — the same build-to-falsify method applied to the finance and AI-control architectures above.
Tools sit below the value proposition. The operating question is what improves: revenue integrity, decision quality, cash visibility, productivity, control or transformation velocity — measured across markets, entities and currencies, not in a single ledger.
Economic impact
Global delivery footprint
Recovery figures reflect solution value architected against identified revenue leakage, not realised cash collected. Portfolio figures reflect programme scope under governance.
Scroll the matrix horizontally on smaller screens →
Twenty-five years of Finance and O2C transformation compressed into the economic or operating result that mattered.
A transformation architecture should make the business logic clearer, the control stronger and the operating model lighter.
Start with leakage, cash, working capital, control or productivity. Select technology only after the economic mechanism is explicit.
AI can accelerate decisions only when the decision path remains inspectable, governable and recoverable.
Working software exposes architectural assumptions earlier than presentationware. These systems are a way to pressure-test the operating idea before it reaches production scale.
Open to conversations around Strategic Finance Architecture, Enterprise Value Advisory, AI Governance, Business Process Improvement, O2C transformation and senior transformation leadership.