Integration engineer in Tallinn, Estonia. My work sits on the seam between development and operations: systems that move business data between companies, and lately the machinery that decides whether a language model's output is allowed anywhere near them.

Agent harnesses. The interesting part of an LLM system is not the prompt. It is everything around it: what context the model is allowed to see, which tools it can call, what checks its output, and when it is permitted to stop. Right now that is a mapping-automation pipeline that generates XSLT transformations and refuses to put any of them in front of a reviewer until they pass a golden-pair correctness oracle. The model writes the code. The harness decides whether that code ships.

MCP servers. Custom servers for GitLab, Jira and internal knowledge systems, and one that drives live WordPress and Elementor sites over a bridge plugin. Designing tools for an agent is API design for a consumer that will not read your documentation and will not ask a question when it gets confused — the schema, the errors and the defaults have to carry all of it.

Integration systems. EDI and API pipelines moving high volumes of business data across global supply chains: EDIFACT, ANSI X12, XML/XSLT/XSD, AS2, SFTP. This is the domain that taught me to care about verification. A wrong transformation is not a failing test, it is a wrong invoice.

Developer tooling. A VS Code extension on the Language Model API with two execution paths — a deterministic one-shot and a separate agentic loop — plus the observability needed to tell which of them failed, and why.

• Correctness has to be machine-checkable, or the agent is just producing confident text.
• Context is a budget. Most agent failures are retrieval and scope failures wearing a reasoning costume.
• A verification loop nobody can read is not verification.

• Languages — TypeScript, Python, Rust, PHP, Bash, SQL, XSLT
• Agents & AI — MCP, VS Code LM API, LangChain (Deep Agents), CrewAI, AutoGen, Langfuse
• Integration — EDIFACT, ANSI X12, XML/XSD, REST, AS2, SFTP
• Web — React, NestJS, Node.js, WordPress, WooCommerce
• Infrastructure — Linux, nginx, MySQL, Cloudflare, Plesk, AWS, GitHub Actions
• Quality & observability — golden-pair validation, XSLT unit testing, SonarQube, Elasticsearch, Kibana, Grafana

Outside the day job I run websculpt.ee, where small businesses get the same discipline applied to much smaller systems:

• Sites and shops — WordPress and WooCommerce builds, Elementor, product catalogues, payments and shipping
• Migrations — moving sites and mailboxes between hosts, with a rollback plan and minimal downtime
• DNS, Cloudflare and certificates — zones, SSL, caching and WAF rules that do not break checkout
• Automation — the weekly manual task, turned into something that runs on its own

Also Arduino and small hardware projects, for the pleasure of a problem that fails visibly.

Always up for a conversation with engineers working where systems integration meets AI automation.