• 21 years in tech (C/C++ → Java/Scala → Python → AI/Agent) — roughly 4 careers in dog years
• Specializing in making AI understand what you meant, and making legacy systems understand it's time
• Working with Tier-1 European banks on the two problems that quietly block enterprise AI
  • Context you can't Trust.
  • Estates too brittle to Modernize.

📍 London, UK · LinkedIn

• Analytics Estate Exit (SAS)
  Migrating SAS workloads — Base SAS, SAS/STAT, DI Studio, Model Manager, Visual Analytics, Viya, and AML suites — to Databricks and Snowflake via automated transpilation and semantic uplift. Statistical logic preserved; platform lock-in eliminated.

• EDW Exit (Teradata · Netezza · Oracle · MS SQL)
  Large-scale EDW migration to Databricks and Snowflake — schema translation, workload profiling, query optimization, and medallion architecture design. The on-premise performance tax: cancelled.

• ETL & Integration Modernization (DataStage · Informatica · ODI · Talend)
  Retiring legacy ETL and re-engineering pipelines as cloud-native data products — embedded data quality, machine-readable lineage contracts, observable orchestration. Opaque batch jobs not invited.

• Mainframe & Batch Orchestration Exit (COBOL · JCL · AutoSys · Control-M · OPC · UC4)
  Extracting decades of embedded business logic from mainframe-bound workloads into testable, versioned, agent-consumable services — SLA contracts preserved, full observability gained.

• Regulatory Reporting & Data Lineage (RegTech)
  BCBS 239, Basel III/IV, FRTB, CRR3, COREP — modernized with machine-readable lineage contracts that surface hidden dependencies before migration, satisfy Principle 2 automatically, and produce data assets trustworthy enough for an agent to reason over. Because regulators don't accept "it was like that when we got here".

• Context Engineering
  Designing context supply chains for enterprise AI — semantic models, data contracts, knowledge graphs, and retrieval pipelines that give agents accurate, governed, auditable answers. Garbage in, hallucination out. This is the fix.

• Composite semantic layers — coordinating semantically-driven artifacts across a heterogeneous estate rather than chasing the universal semantic layer that never arrives; OSI compliance as a non-negotiable evaluation criterion for any target architecture from 2026 onwards

• Semantic portability — designing exit-readiness into vendor-managed semantic layers before migration commits, not after. Snowflake Cortex, Databricks Unity Catalog, and Google Agentic Data Cloud all have the same lock-in surface as the platforms they replaced. New labels, familiar trap

• Semantic observability — detecting semantic drift in agent-interpreted business logic before it surfaces as a Principle 2 breach, not after the auditor asks why the number changed. Auditors do not accept "the model seemed confident"

• Semantic-first agentic AI — agent-ready target architectures where governed semantic models are the query surface, not raw tables; the difference between a migration that enables AI and one that merely moves the problem to the cloud at three times the cost

• MCP + A2A — vertical agent-to-tool and horizontal agent-to-agent protocol adoption in regulated estates; where the governance and identity model sits across that stack in a BCBS 239-compliant architecture. Ignoring both is just vendor lock-in with better branding

Databricks Snowflake GCP Python SQL Data Mesh Knowledge Graphs