Jere Kelonen — Engineering lead in Helsinki, working on applied AI
I build software on top of how companies actually run. Since 2019 that's meant taking processes out of the ERP and turning them into working technical solutions, like CAP applications on SAP BTP, for companies in many industries, among them manufacturing, forestry, logistics, shipbuilding and food. More and more of it now is LLM‑based software running in production inside those processes.
I ended up in enterprise software by accident. After my internship at Sanoma I wanted to go toward AI, but my first real developer job happened to be at an SAP company. The interest never went away, so when the first GitHub Copilot came out I was all over it, and I've been building with these tools ever since.
Day to day, my work covers LLM-based features in production software, building with coding agents, full-stack TypeScript and Python from the data model to the UI, cloud architecture, and working closely with a team of senior engineers and architects.
The part I enjoy most is when an idea is still rough. Talking it through until it has a shape, then pushing it until it runs, and then until it scales.
Understand the thing.Build the thing.Repeat.
A few recent projects. Client work stays with clients, so these are described in general terms.
An AI system that reads unstructured incoming requests, turns them into structured data with foundation models, and routes the work by complexity. Most of the engineering went into what surrounds the model: an audit trail, a kill switch, and concurrency handling that holds up as volume grows.
An application that shows which company in a corporate group uses what share of the group's cloud platform cost, a split the platform's own reporting can't make. It collects usage and cost data, applies allocation rules that administrators maintain, and shows controllers the result. Taken from the domain model to production with strict test‑driven development.
A full-stack platform for supply-chain planners, with a domain-driven backend that coordinates several external systems and a set of integrated frontends. Planners configure and run the statistical calculations themselves.
A failover architecture for business-critical cloud applications: the landscape analysis behind the design, a secondary region on another hyperscaler, and automatic traffic redirection, so a regional outage doesn't take the applications down.
