AI Engineer Backend (Remote - ES/UK only)

About Us Quadrivia is the health technology company behind Q, a comprehensive, controllable, and customizable assistant AI built by clinicians, for clinicians. Addressing the urgent shortage of healthcare professionals, Q provides real-time, personal, and reliable support for clinical tasks across the care continuum. Designed for providers, payers, and pharmaceutical companies, Q is easy to customize and integrates seamlessly into workflows, delivering precise assistance across the care spectrum. The Role You'll build and run Cortex , the core AI architecture behind Qu, and the services that sit on top of it: automated AI audits, patient simulators, retrieval (RAG), and the escalation agents that take over in red-guardrail situations. This is a backend role first. The job is to make our AI systems reliable, fast, and observable in production, not to invent new ML. You own the software underneath the agents. We're a small team that ships real systems. We've built our entire AI-driven evaluation system, our voice orchestrator, and a multi-hierarchical RAG platform from scratch. What You'll Do * Design and maintain robust, modular backend systems using clean architectural (SOLID) principles to ensure long-term maintainability, scalability and flexibility as the agentic stack evolves. * Own Cortex end-to-end: architecture, API design, service boundaries, reliability targets, and proactively managing failure modes. * Build the platform services around it. The automated audit and eval pipeline, patient simulators for testing agents at scale, and the retrieval layer. * Write fast, well-tested Python services with FastAPI, asyncio, and pydantic, and get the queues, caching, and data stores right. * Wire up the multi-agent orchestration: routing between agents, shared state, and clean tool interfaces. * Engineer the RAG pipeline for high-signal retrieval (chunking, hybrid search, re-ranking, caching) and prove the grounding holds. * Make the whole thing observable: structured logs, OTEL tracing across the agent graph, cost, latency and token visibility, dashboards, and CI gates that catch regressions before they ship. Minimum Qualifications * Your core is backend and software engineering. You write clean, maintainable services and you care how they behave in production. * Deep understanding of architectural design patterns (e.g., Clean/Hexagonal Architecture, Domain-Driven Design, SOLID, event-driven) to manage complex system boundaries. * At least 2 years, demonstrable, building or scaling user-facing AI software that real users touched. We'll want to see it. * Expert Python, with strong FastAPI, asyncio, pydantic, and production observability. * Comfortable with agent patterns and eval-driven development. * You've worked at a startup before and know what wearing several hats actually costs. Nice to Have * Real-time and voice: WebRTC, LiveKit, SIP, VAD, barge-in, turn-taking. Useful here, not required. * Programmatic prompt optimization techniques. * LLM-as-judge setups and other evaluation tooling. * GCP: Cloud Run or GKE, Pub/Sub, Vertex AI, GCS, Secret Manager, Cloud Logging and Trace. * Healthcare data familiarity. Example Problems You'll Tackle * Stand up the AI audit pipeline so evals run automatically on slices of production traffic, with regression gates wired into CI. * Build a patient simulator that lets us stress-test agents at scale before they ever reach a real call. * Improve the RAG pipeline with hybrid retrieval and re-ranking, then prove the gains with faithfulness and context metrics. * Get OTEL-first tracing across the agent graph, with automated eval triggers on live traffic. * Turn EHR integrations into reliable tools the agents can call. Tech Stack Python, FastAPI, pydantic, asyncio, Redis, Postgres, vector stores, Docker, Kubernetes, Terraform, ArgoCD, OTEL, TypeScript, React. Real-time stacks (WebRTC, LiveKit, SIP, STT/TTS) where the work touches voice. What Success Looks Like * Quadrivia's backend becomes a reference for reliability, safety, and performance. * Your services run above 99.99% availability under strict regulatory constraints. * Other engineers build new clinical workflows and agent capabilities quickly and safely. * AI-generated code gets reviewed, corrected, and owned by you.