Solutions Architect / Fullstack Developer: Agentic Interfaces & AI

Solutions Architect / Fullstack Developer: Agentic Interfaces & AI Location: Malmö, Sweden About the assignment We're building an agentic runtime platform for multiple customers (multitenant) for advanced video systems. An LLM is one selectable execution layer alongside deterministic code, with device level perception running on cameras and tool servers. The interaction design is still experimental. We're looking for someone who can implement it and find the right concept. You will define how users understand, control, moderate, and trust agents acting against real infrastructure: cameras, recording devices, body worn devices, where a faulty action can have physical, legal, or ethical consequences. You'll write the interaction contract between the runtime platform and its frontend. What you'll build The client/runtime protocol. We believe AG UI will be central. You'll own the conformance specification: event mapping, shared state schema, reconnect and replay semantics, and define which changes to the runtime environment are required to meet the spec. Moderated flows. Human in the loop approval for actions with significant consequences: approve, reject, edit, standing consent, and post hoc review. Must work regardless of whether the call originates from a model, a sandboxed script, a sub agent, or a remote agent, including when no human is present. Guided flows. Context aware entry points, "plan first, confirm later" workflows, and schema driven clarification handling. An empty prompt box facing a large federated tool surface is a failure state, not a product. Clarity in a hybrid execution model. Users must always know whether a result came from reasoning, perception, or computation, and with what confidence, cost, and latency, even when no model was involved. It should also be possible to turn such a run into a deterministic, saved script. Security as part of the user experience. Policy denials, failed tool authorizations, tenant boundaries, and untrusted tool output must be visible and understandable, never silently absent. We believe this is about an interface that's 100% dynamic The system should build its interface based on what decisions the AI makes from the camera feeds, not a static UI cluttered with buttons. You'll need technical depth, but also a genuine feel for interaction design and the "feeling" of the interface. Who we're looking for A creative problem solver with broad competence: a solutions architect and fullstack developer with a strong weighting toward AI. You're deeply versed in the AI world and thrive in projects where we know what we want to achieve, but not always exactly how, or whether it's even feasible. You're driven, proactive, and resilient. A thick skin matters, as does the ability to engage directly with stakeholders. Must have At least 5 years of fullstack development experience with a frontend weighting Hands on experience with TypeScript and React Experience designing and implementing microfrontends or independently deployable modules Driven and proactive, comfortable engaging directly with stakeholders Strong Python and AI skills. This is the heaviest weighting Nice to have Experience with AG UI (a big plus), MCP, A2A, or other event driven protocols Experience defining protocols and interaction contracts Backend development in Python Design for uncertainty and AI driven outcomes Human in the loop solutions Security in AI and multitenant systems Fluent UI v9 or an equivalent design system Streaming based user flows and state management Practical details Location: Malmö, Sweden Start: as soon as possible Duration: at least 6 months, with the overall project running for several years Onsite at the client at least 4 days per week