DocLine: Agentic Healthcare Triage for the NHS

The NHS's 111 non-urgent line takes over 22 million calls a year, with average waits of 30 minutes, judgement calls made under time pressure, and legacy systems that can't share patient history across a call. DocLine replaces this outdated, manual process with a coordinated team of AI agents. Band enables multi-agent collaboration that patients are passed through - agents are recruited into a case live as their skills are required, while the framework utilises Band's persistent-chat-history to ensure that each patient is treated with all their medical records and past conversations under consideration. Each agent is wired to databases and skill files to ensure that every agent follows both their individual protocols, as well as NHS compliance guidelines. AI/ML API allows DocLine to call varying models with frontier-grade reasoning to execute these high-stakes tasks, while the higher-volume conversational work runs on Featherless AI, allowing us to match our compute needs efficiently. The combination of these technologies, alongside our multi-agent framework, creates a system where agents are able to instantly respond to non-urgent care patients and provide all the necessary support, while doctors are able to do their jobs and the NHS can dedicate its funding efforts to improving the quality of its medical infrastructure. Most multi-agent systems hardcode a fixed sequence between agents, so the same pipeline runs regardless of the patient in front of it. DocLine's architecture is built to behave differently: a specialist agent is only recruited into a case once it has actually been flagged as high-risk, meaning the system adapts its own structure to the severity of the call rather than forcing every patient through an identical chain of agents. This dynamic recruitment, paired with Band's persistent history rooms, is what lets DocLine treat a one-off cold as routine while still catching the rare, escalating case that a static pipeline would miss.