About this role
Three concrete problems on our plate: Work instructions from video. A process engineer records a short video of an operation; our system produces a structured, version-controlled instruction the team can review and ship. A continuous-improvement agent. Dashboards already show deviations. We want an agent that proposes evidence-backed instruction changes: not "cycle time is up on station 3" but "rework concentrates on step 7, here is the fix and the data behind it." Multi-level BOMs and nested assemblies. Industrial products are deeply nested; we handle the leaves but not the tree. Data model, variant rollup, integration with engineering data, routing. Foundational, and what unlocks our most variant-heavy customers. The stack: Our core stack today is Node.js , TypeScript, React, and MongoDB on cloud infrastructure. The upcoming work stretches it significantly: LLM and agent orchestration, MCP-style tool use, computer vision for shop-floor verification, video understanding for instruction generation, software interfaces for robots and humanoids on the factory floor, observability and deployment across multi-tenant SaaS, private cloud, and on-prem / air-gapped factories. Strong experience in Java, .NET, Go, Python or similar is fine. Mindset matters more than prior framework experience. Learning a stack is easier than learning how to think about software at scale. About our platform We are building the execution platform for hybrid manufacturing : the system that orchestrates work on the shop floor across humans, AI agents, and robotic systems . The platform is used daily by thousands of operators across Europe to design, execute, and continuously improve work in high-mix, low-volume factories. The technical reality is rich: multi-level BOMs, variant-heavy product structures, sequence-dependent execution, version-controlled executable work. And the bet on top is bigger: vision-based step verification, agentic execution supervision, autonomous continuous improvement, factory replay for model training. None of this is slideware. It is what we are building over the coming years.