The role sits in the ML Platform (TIGER) team within AI & Machine Learning.
The goal is a platform where each feature team defines an agent once and gets a production-ready, secure and compliant agent automatically.
The platform underpins Consultation Assistant, Phone Assistant, Health Companion and other products, with a shared agent runtime, a unified GenAI gateway, and observability and evaluation tooling aimed at more reliable AI for patients and care teams across Europe.
The tech team builds products and features meant to make daily work easier for care teams and patients.
The company runs one fully cloud-native platform with web and mobile interfaces, uses Rails, TypeScript, Java, Python, Kotlin, Swift and React Native, and applies AI ethically across its products.
Responsibilities
Design the agent manifest and registry: one declarative spec covering skills, tools, memory, model, guardrails, sandboxing and security, which connects any new agent to the full platform. Agents can be triggered from product surfaces, other agents via A2A, schedules or workflows.
Run the GenAI gateway built on LiteLLM end to end, covering routing (including KV-cache-aware routing), failover, metering, cost attribution, reserved-capacity pooling and guardrails across self-hosted vLLM and managed providers, at several hundred million requests per month.
Build reliable infrastructure for long-running, streaming agent sessions: SSE and WebSocket connections, queues, retries, human approvals, and snapshot and resume, inside a sandbox consistent with the existing application runtime.
Build feature-flagged progressive rollouts for prompts, models and configs, including an emergency kill switch and an audit trail for agents acting on sensitive health data.
Own reliability for the agents platform: define SLOs, lead on-call and incident response, and write runbooks, including provider failover for the GenAI gateway.
Build the observability and evaluation layer: online traces and logs, offline evals (LLM-as-judge plus patient and practitioner simulators) that gate CI, and a compliance-grade audit trail for health data.
Ship one or two reference agents yourself before product teams adopt the platform, and advise feature teams on agent architecture trade-offs.
Requirements
If you don't match this profile exactly but believe the role fits your skills, you are still encouraged to apply.
5+ years of software engineering experience, with hands-on production work building or operating LLM-based agent systems (Google ADK or similar) and a solid grasp of planning and reflection loops, tool calling, state and memory management, multi-agent coordination and human-in-the-loop checkpoints.
Experience with agent interoperability and tool protocols (MCP, A2A, AG-UI), control-plane design for asynchronous or long-running systems (queues, retries, checkpointing), and feature flags or progressive rollout with reliable kill switches.
Experience managing LLM cost and capacity: routing, caching, and reserved-capacity planning with managed model providers.
Ability to own production reliability (SLOs, incident response, on-call), with a security mindset for sandboxing, permission scoping and audit trails for autonomous systems that handle sensitive data.
Excellent communication and collaboration, high agency, a pragmatic build-versus-buy approach, and comfort in a founding role where shaping scope is part of the work.
Fluent English.
Bonus: experience giving agents richer identity, memory and knowledge (long-term memory, artifact stores or RAG pipelines), ideally alongside a data platform team, or building a feedback loop that turns production traces into evaluation datasets.
Bonus: experience with Kubernetes-native abstractions (CRDs/operators) for declarative agent lifecycles, and/or durable stateful execution engines such as Temporal.
Bonus: experience in regulated environments (healthcare, finance, public sector) and with frameworks such as GDPR, HDS, ISO 27001 or SOC 2.
Stack: Python, plus TypeScript or Go for agent SDKs and tool integrations; GCP, AWS or Azure; Google ADK running as services on the application platform; MCP, A2A and AG-UI; Flagd or similar for flags and kill switches; LiteLLM; reserved capacity such as Azure OpenAI PTUs, Vertex AI GSUs or AWS Bedrock Provisioned Throughput; MLflow; Braintrust and Datadog LLM Observability; Kubernetes (GKE or EKS), Docker, Helm, Kustomize; Terraform; Datadog. Other relevant tools are welcome.
Benefits
Comprehensive basic health insuration for you and your children, at no cost to you
25 days of paid vacation per year, plus up to 14 days of RTT
Free mental health and coaching services through partner Moka.care
Up to 10 days per year of working abroad under the flexibility days policy
Swile lunch vouchers worth €8.50 per working day, with €4.50 paid by the company
A work council subsidy that refunds part of a sports club or creative class membership
50% reimbursement of public transport subscriptions
ParentCare Program: 100% salary during the first month of birth leave and 75% during the second, paid by the company
Enrollment in the long-term employee value sharing plan DoctoGrowth
For caregivers and workers with disabilities: remote policy adjustments, extra days off for medical reasons, and psychological support
Relocation support for international moves
Access to top AI tools for coding and development, plus dedicated training