Torsten Schippel · Java Enterprise · AI Agents
I bring AI agents into existing Java systems — without a rewrite, without a platform switch. 20+ years of enterprise architecture, since 2024 consistently on the JVM with LangChain4j, Spring AI and MCP. Java Enterprise since 1996, today AI agents on the JVM.
The core idea
Transactions, idempotency, audit, permissions, human approval. That is where an agent becomes production-ready — and where most agent frameworks lack experience from classical enterprise systems.
About
I have been writing Java since 1996. For eight of those years I modernized manufacturing execution systems at Dürr Systems: JBoss to WildFly, stored procedures to Java, monoliths into containers. Before that, core banking at GAD, and before that, PKI and smartcards. These systems were never greenfield. They had to keep running while they changed.
Since 2024 my focus has been on agents. Not on the next chat window, but on the question of why agents fail in production: they know no transactions, no idempotency, no audit trail, no approval. That is not an AI problem. It is the problem I have worked on for twenty years.
That is why I build on the JVM. Not because Java is new, but because that is where the systems that remain are running. Where Java Enterprise meets AI Agents: the existing system grows, it is not replaced.
I work as an individual, not as an agency. Short paths, direct communication, code instead of slide decks. On-site in northern Germany, otherwise remote.
I develop with AI support (Claude Code, OpenCode, GitHub Copilot) — but with human responsibility: architecture, review and code quality remain with me.
Torsten Schippel · Hamburg
Experience
Professional stations from core banking and MES to Agentic AI.
Own projects
Not a repository catalogue. Two own projects that together form a platform: the agent runtime and the three stages of orchestration.
Own reference projects and GitHub repos — not client work.
Top project
QOCThe runtime that makes an agent runnable.
ProblemAgent demos stop at the surface. Sessions, tools, permissions and persistence remain DIY work.
Java/Quarkus runtime for self-hosted coding agents. The OpenCode migration brings sessions, context, tools, permissions, MCP and SSE streaming to the JVM — with SQLite or PostgreSQL. The core is pure Java, without Quarkus, Jackson or HTTP.
Repositoryagentic-playgroundOne platform in three stages.
ProblemAgent demos stop at runtime. Orchestration, durability and replay are left out.
Monorepo that orchestrates the same LangChain4j agent across three stages: Quarkus in-process, Kafka event-driven with Postgres and Redis, Temporal durable. The agent stays the same — only the orchestration moves outward.
Proof
Plus: PostgreSQL, Redis, MongoDB, Neo4j, Docker, Kubernetes, AWS, GitHub Actions. Python and TypeScript where they fit.
Contact
You have a Java system and an idea of what AI could do inside it? Send me a short message about what it is about. I will reply with an honest assessment — even if I am not the right fit.
I am also available for regular development projects — with AI support (Claude Code, OpenCode, GitHub Copilot) for code reviews, refactoring, tests and documentation.