Work

Challenge: deliver a modern GCP data platform without breaking the on-prem estate the bank still reports from. Both roadmaps had to live in the same backlog, serving CDO, Risk, Finance, Treasury, Corporate Bank, and Private Bank simultaneously.

The problem

• A mission-critical on-prem data estate was still the system of record for regulatory submissions — it could not go dark during migration.
• Six divisions each had distinct data demand, governance expectations, and delivery cadence, with no single shared roadmap.
• BCBS 239 and GDPR controls had to be embedded in the platform itself, not bolted on as downstream checks.

What I did

• Owned platform product strategy and lifecycle across multiple scrum teams, sequencing cloud migration as low-risk increments tied to business value rather than a big-bang cutover.
• Built the GCP architecture (BigQuery, Dataplex, Cloud Composer) in parallel with the legacy estate, keeping both audited and compliant.
• Ran cross-divisional demand management so each division got a roadmap meeting them where they were, against one shared platform.
• Made compliance an acceptance criterion — regulatory accuracy designed into the pipeline, not inspected after the fact.

Read the full case study →

Challenge: data stewardship across compliance pipelines was manual, slow, and inconsistent. Business rules and metadata were authored by hand, creating a bottleneck that scaled with the data, not with the team.

What I did

• Redesigned the stewardship workflow to be API-led, inserting Generative AI (Vertex AI) for business-rule generation and automated metadata enrichment.
• Kept a human-in-the-loop review on generated rules so accuracy and auditability were never traded for speed — essential in a regulated domain.
• Measured quality gains alongside effort reduction, so the automation case rested on output quality, not just throughput.

The lesson

The win wasn't replacing stewards — it was removing the repetitive authoring so they could spend their judgement where it mattered. Production reliability over demo polish was the design principle throughout.

Read the full case study →

Challenge: deliver a cloud-native content and market data platform at institutional scale, while replacing a large manual content operation with automation that analysts could actually trust.

What I did

• Led discovery across 260+ products to define platform strategy and a prioritised backlog.
• Defined and delivered the AWS platform serving 1,000+ institutional users across 35+ markets.
• Built an ML/NLP content extraction pipeline pulling 250+ attributes from raw feeds across 60+ language pairs.
• Scaled content operations from 208 manual specialists to 10 — a 95% reduction — without losing coverage.

Read the full case study →

Challenge: the brief called for a separate compliance portal — a post-submission check advisors would visit after building a recommendation. Discovery showed advisors described it as "another system I'll check eventually," which broke the compliance objective before it shipped.

What I did

• Reframed the constraint: the rule wasn't "compliance must interrupt the flow," it was "compliance must happen before the recommendation is made." That distinction unlocked an inline suitability architecture — the check runs in the background as the advisor builds the recommendation and surfaces the result exactly when needed.
• Built suitability scoring across Structured Products, Mutual Funds, ETFs, Private Equity, and Hedge Funds for UHNW, HNW, Corporate, and Trust clients.
• Delivered the Front Office / Middle Office target operating model, resolving a long-standing architectural gap in cross-border client onboarding.
• Led the Brexit legal-entity migration (UK to Luxembourg): locked acceptance criteria with UK compliance and Luxembourg entity leadership before any code was written, then ran a two-week parallel-run window with both entity structures live simultaneously.

Read the full case study →

Why: the CCO sign-off pattern from Deutsche Bank — explainability, audit trail, human-in-the-loop, confidence gating — isn't a one-bank trick. Between roles, I've been proving it travels by building it as working code, not slides.

What's live

• Model Risk Copilot — SR 11-7 / SS1/23 model validation: documentation analysis, compliance-gap scoring, auto-drafted validation reports.
• KYC/AML Intelligence Agent and AML Copilot — entity screening, adverse media triage, SAR narrative drafting with a confidence gate before anything reaches a human reviewer.
• KPI Integrity Agent — anomaly detection that checks the data quality behind a metric spike before trusting the signal, not just whether the number moved.
• Data Intelligence Platform — the metadata, lineage, and DAMA-DMBOK data quality agents this whole pattern is built from, extracted from the Deutsche Bank CDO catalogue design.
• Responsible AI for FinServ — the shared controls library (ConfidenceGate, AuditChain, HITLQueue, DriftMonitor, ExplainabilityWrapper) every other repo imports.

Read the full case study →