Architecture katas.
Practice exercises, kept apart from my production projects. Each starts from a brief with hard constraints and ends in a full design: diagrams, component rationale and open questions.
These are design studies, not delivered systems. The figures in them are design assumptions, not measurements.
Reliable data
Data that can be shown to be complete, consistent and traceable.
K—01
Featured
Cold-Chain Telemetry and Compliance Platform
20,000 refrigerated vehicles, 72-hour offline gaps, and GDP reports that an auditor can verify instead of having to trust.
Sector
Pharmaceutical logistics
K—02
Featured
Open Lakehouse with Data Contracts and a Shared Semantic Layer
Six brands keep their query engines, shared data is written once as an open table, guarded by a contract in CI and one metric definition.
Sector
E-commerce group
K—03
Regional Health Information Exchange
Eight EHR systems into one FHIR record per patient, reversible patient matching, consent on every read, ten-year audit trail.
Sector
Healthcare
K—04
Enterprise Data Warehouse Migration to BigQuery
Move a 60 TB SQL Server warehouse and 1,900 ETL packages to BigQuery in 12 months, wave by wave, with every table reconciled.
Sector
Retail data platform
Decisions
One answer to one question, within a latency or cost budget.
K—05
Real-Time Fraud Decisioning for Instant Payments
Score every instant payment for fraud in 150 ms, keep each decision reproducible for seven years, touch the mainframe only through MQ.
Sector
Banking
K—06
BSS Modernisation and Real-Time Customer 360
Build a Customer 360 from change events of 14 systems without touching the mainframe, then retire billing by cohort over five years.
Sector
Telecommunications
K—07
Unified Marketing Intelligence on Google Cloud
One certified metric layer over a GA4 stream, four batch sources and the warehouse, so dashboards, an LLM and agents quote one number.
Sector
E-commerce marketing
K—08
Competitor Price and Markdown Intelligence Platform
Collect competitor pages within legal limits, match them to 120,000 SKUs with measured precision, advise on markdown timing with evidence.
Sector
Retail pricing
AI & GenAI
Models recommend under governance; rules and people decide.
K—09
Featured
Governed GenAI Data Agents with LLM Evaluation
Three enterprise agents limited to certified metrics, contracted data and permitted documents, with traced answers and gated releases.
Sector
Enterprise analytics and AI governance
K—10
Claims Intake with Document Intelligence
Self-hosted models recommend, rules and adjusters decide, eight GPUs sit behind a queue, every decision is auditable for seven years.
Sector
Insurance
Facing a design problem like these?