Member of Technical Staff - Vision Agents

The Role

• 5C is building agentic systems for radiology - long-horizon reasoning over imaging studies, structured evidence ledgers, triggered specialists, and constrained report synthesis. The agents themselves are the work of the modelling and clinical teams. The harness around them is this role.
• We are hiring a Member of Technical Staff - Vision Agents whose primary job is to build the harnesses, evaluation infrastructure, and orchestration around the vision agents we ship. Agents are only as good as the harness they run inside. This role owns the harness.
• We are an applied AI team, not a research lab. Your output is reproducible eval suites, production-grade orchestration, and the telemetry that turns agent behaviour from opaque to debuggable. Papers are a side-effect of the work, not the work.
• We are strongly biased toward candidates who can join quickly. If your notice period is short, your application will move faster because these teams are being built now.

What You Will Build

• Agent harnesses: orchestration layer, tool integration, structured state passing, retry and fallback logic, deterministic replay, and cost / latency governors
• Evaluation infrastructure: reproducible test suites, golden case sets, regression catches, A/B comparators across model versions, calibration and uncertainty surfaces
• Observability and tracing: span-level visibility into agent steps, prompt rendering, tool calls, and intermediate reasoning - built so a radiologist or a model owner can debug a failure mode in minutes, not hours
• Trajectory tooling: replay any production run, intervene at any step, extract preference / RLHF data from real radiologist edits, and feed it back into the next training cycle
• Production deployment of agents at clinical reliability - 24/7 uptime, deterministic behaviour under load, graceful degradation when an upstream model is unavailable
• Tight loops with the modelling team and the radiologists who use the system. The harness is what makes their feedback actionable

You Should Have

• 2+ years of hands-on experience building production systems on top of LLMs or multi-step AI workflows - not just calling APIs from a notebook
• Strong Python and (ideally) TypeScript, including async, structured data plumbing, and serious testing instincts
• An eval-first mindset: you instinctively reach for reproducible measurement before changing a prompt, a model, or a tool
• Experience with at least one of: agent frameworks (LangGraph, DSPy, custom), eval frameworks (Inspect, Braintrust, OpenAI evals, custom), tracing tools (Langfuse, OpenTelemetry), or RLHF / preference data pipelines
• Clear thinking about what makes an agent system fail in production: distribution shift, tool errors, latent loops, hallucinated tool calls, silent regressions

Even Better If You Have

• Built or contributed to an open-source eval framework, agent harness, or tracing library
• Worked on agentic systems in a high-stakes domain - clinical, financial, legal, or critical infrastructure
• Strong opinions on the line between research and engineering in applied AI, and the maturity to hold both honestly
• Published a postmortem of an agent failure mode that you actually fixed

Why 5C

• 5C is one of the best places to work for serious, engaged builders. We are a leading company in agent-driven software development and a leading business in AI for medical imaging at the same time
• You will work on systems that move from code to clinical impact unusually quickly: reports delivered faster, radiologists made more effective, hospitals able to serve more patients
• Resumes are not important. Proof of work and proof of a high learning rate matter more. Show us systems you built, products you shipped, open-source work, customer outcomes, or unusually strong side projects
• We are strongly biased toward candidates who can join quickly. If your notice period is short, your application will move faster because these teams are being built now.