AI‑assisted engineering and architecture

• Back-end services and APIs

  Java and Spring Boot, REST APIs, PostgreSQL, messaging on Kafka and RabbitMQ, integration and process automation on Camunda, deployed with Docker and Kubernetes on AWS or GCP.

• Legacy code and modernisation

  Characterisation tests first, drafted by agents and reviewed, then refactoring, framework and Java upgrades, and extracting services. A target architecture and a migration path that keeps the business running: strangler-fig extraction, parallel runs, no big-bang cut-over.

• Architecture assessment and documentation

  A review of an existing system against the quality goals that matter to you: performance, reliability, security, maintainability, cost. You get the risks, the trade-offs and a prioritised list of fixes, in an arc42 document with C4 diagrams and a log of decision records. Agents draft from the code and our conversations; I check every claim against the running system.

  Fixed scope. The usual place to start.

• AI-assisted development for your team

  Set your team up to work with coding agents: CLAUDE.md or AGENTS.md per repository, tied to your conventions; skills and hooks for ticket kickoff, naming checks and standards review; Jira and Confluence specs over MCP, so the agent reads the spec before it writes code; and secrets kept out of what the agent can see.

  Fits when the team has started using agents and the results depend on who is prompting.

• AI features in your product

  LLM and agent features: model API integration, tool calling and MCP servers, with the authentication and access control they need.

• Fractional architect

  A few hours a week to review designs, guide teams and keep decisions consistent across services. Includes cloud cost review.

Where I use it

A telecom operator’s customer identity platform, in a major Keycloak migration now in testing: data and session migration scripts, test suites and the new proxy. A SaaS company’s identity application, APIs and OpenFGA model. A Norwegian ERP software company, where I set up the team’s Claude Code workflow. And keycloak-otp, where 25 of 44 commits are co-authored with Claude Code.

What the agents do

Draft code, tests, Keycloak configuration, migration scripts and Java extensions. Read an unfamiliar codebase quickly. Draft decision records, arc42 sections and C4 diagrams. Give a first-pass review of code and designs.

What stays with me

Design and security decisions, a line-by-line review of everything that ships, and responsibility for the result.

Guardrails

Tests with every change, static analysis and dependency scanning in CI, small pull requests.

Your rules

I follow your policy on which tools may see your code, use your approved accounts where you have them, and work without agents where you ask. Secrets and real user data stay out of prompts.

ISO/IEC/IEEE 42010

What an architecture description must cover: stakeholders, their concerns, and the views that answer them.

arc42

The structure for the documentation, from goals and constraints to decisions and risks.

C4 model

Diagrams at fixed levels of zoom: system context, containers, components.

Architecture Decision Records

One short record per decision: the context, the decision, and its consequences.

3 services

extracted from a monolith over 6–12 months, with no planned downtime.SaaS company

Around 10–15%

less cloud spend, from rightsizing and optimising applications and data processing.SaaS company

15 days to 25 minutes

for one business process, after automating it.Telecom operator

2+ of 3+

proofs of concept on microservices and BPMN orchestration taken to production, in a legacy modernisation.Telecom operator

3 extensions

written in Java for Keycloak: an authentication flow, themes and an event listener.Web platform operator

Claude Code, Codex, Java, Spring Boot, REST, PostgreSQL, SQL Server, Kafka, RabbitMQ, Camunda / BPMN, Docker, Kubernetes, AWS, GCP. Oracle Certified Professional, Java SE 8. Camunda Certified Engineer.