Role overview
Develop the software and operational foundations for a real-time voice AI platform used by healthcare professionals to configure, monitor and manage clinical agents. The work spans agent architecture, evaluation, security, integrations and the supporting web application, with a focus on reliable operation as conversation volumes grow.
Responsibilities
- Design the agent runtime, including tool registration, token budgets, stopping rules, context handling and retry behaviour.
- Choose multi-agent patterns and build planning workflows with saved state, checkpoints and human review before consequential patient-data actions.
- Create agent skills and integration servers for hospital and electronic health record systems, with access controls and traceable reads and writes.
- Build evaluation systems with regression data, measurable criteria and automated checks for quality and security risks.
- Add tracing and monitoring for cost, latency, step success, transcription quality and cache performance.
- Develop the clinician-facing web application, improve backend infrastructure and automate deployments with Docker and CI pipelines.
Requirements
- At least three years building production software, including at least one year delivering LLM systems that use tools and state and handle real operational failures.
- Strong Python and TypeScript skills, with the judgment to review and reject code, including generated code.
- Experience designing validated, typed agent tools, managing context and memory, and implementing MCP servers with authorization.
- An evaluation-led approach to non-deterministic systems, including defining metrics and datasets and using an evaluation framework.
- Experience with relational data models and authentication and authorization flows.
- Security judgment for agent systems, including least privilege, indirect prompt injection, execution isolation and sensitive data handling; solid Docker and CI knowledge.
- Fluent spoken and written Spanish and English, strong communication and autonomy, and residence in Valencia or Madrid.
Nice to have
- Applied model fine-tuning experience, or experience serving inference with tools such as vLLM, SGLang or TensorRT-LLM.
- Experience with clinical speech recognition or voice models, WebSockets or WebRTC, and streaming APIs.
- Public work on MCP servers, agent skills, tools, papers or benchmarks; familiarity with healthcare regulations such as GDPR, MDR or HIPAA.