Senior Agents Platform Engineer (x/f/m)

Set a new pulse for healthcare!

We are looking for a Senior Agents Platform Engineer to join the ML Platform (TIGER) team in AI & Machine Learning .

Your mission will be to build the platform that lets every feature team at Doctolib declare an agent once — and get a production-grade, secure, and compliant AI agent by default. You will work on the foundational layer that powers Consultation Assistant, Phone Assistant, Health Companion, and more: a shared agent runtime, a unified GenAI gateway, and the observability and evaluation infrastructure that makes it all trustworthy — contributing directly to safer, faster, and more reliable AI for patients and care teams across Europe.

Working in the tech team at Doctolib means building innovative products and features to improve the daily lives of care teams and patients.

What you'll do

Your responsibilities include but are not limited to:

Own the agent manifest and registry — design the single declarative specification (skills, tools, memory, model, guardrails, sandbox, security) that wires any new agent into the full platform automatically, triggerable from product surfaces, other agents (A2A), schedules, or workflows

Own the GenAI gateway end to end (built on LiteLLM) — manage routing (including KV-cache-aware routing), failover, metering, cost attribution, reserved-capacity pooling, and guardrails across self-hosted (vLLM) and managed providers, serving several hundred million requests per month

Operate the infrastructure for long-running, streaming agent interactions — build production-grade reliability for SSE/WebSocket connections, queueing and retries, human-in-the-loop approvals, and snapshot/resume, in a sandboxed execution environment consistent with our existing application runtime

Build feature-flagged, progressive rollout for prompts, models, and configs — including an emergency kill-switch and the audit trail that lets agents act safely on sensitive healthcare data

Own reliability for the agents platform end to end — define SLOs, lead on-call and incident response, and build 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) gating CI, and the compliance-grade audit trail for health data

Validate the platform by shipping one or two reference agents yourself before product teams adopt it, and advise feature teams on agent architecture trade-offs

Who you are

Before you read on: if you don't have the exact profile described below, but you feel this job description matches your skill set, we still encourage you to apply.

You'll be a great fit if you:

Have 5+ years of software engineering experience, including hands-on production experience building or operating LLM-based agent systems (Google ADK or similar), with a solid understanding of agent design patterns: planning/reflection loops, tool calling, state and memory management, multi-agent coordination, and human-in-the-loop checkpoints

Have experience with agent interoperability and tool protocols (MCP, A2A, AG-UI), control-plane concepts for asynchronous or long-running systems (queueing, retries, checkpointing), and feature-flagging or progressive-rollout systems including reliable kill-switches

Have experience managing cost and capacity for LLM usage — routing, caching, and reserved-capacity planning with managed model providers

Are comfortable owning production reliability (SLOs, incident response, on-call) and are security-minded when designing sandboxing, permission scoping, and audit trails for autonomous systems acting on sensitive data

Have excellent communication and collaboration skills, high agency, a pragmatic build-vs-buy mindset, and are comfortable working in a founding role where shaping the right scope is part of the job

Are fluent in English

It would be fantastic if you:

Have experience giving agents richer identity, memory, and knowledge (long-term memory, artifact stores, or RAG pipelines), ideally in partnership with a data platform team, or experience building a feedback loop that turns production traces into evaluation datasets