People rarely describe what’s wrong in clear sentences. They circle a worry, start in the middle, understate the issues that frightens them so as not make it real in their heads, and overstate the thing that’s on top of mind. When a person is unwell, attention and clarity are exactly the resources in shortest supply, and across literacy and economic divides, that problem is compounded.
Most inta were built for the opposite of this. They listen for keywords and route on limited ontological matches: “headache” → neurology, “stomach pain” → gastro. A great clinician does something different. They read the whole exchange for what a person actually means, weigh what’s likely against what’s said, and gently fill the gaps the patient couldn’t.
Reading for intent, not keywords
IndRAH treats a conversation the way interpersonal-communication research does: as a balance of social and cognitive work. Understanding someone requires perspective-taking, the work of modelling what they are trying to convey, and an intentionalist read of why they said it that way. A symptom mentioned in passing can matter more than the one stated first, and the job is to notice.
Grounding what’s said against what’s plausible
Borrowing from predictive-coding accounts of perception, IndRAH’s agentic workflows hold an evolving expectation of where a clinical story is going without leading the patient, our agents are built to update its decision paths against each new utterance. That lets a half-said symptom survive, and an over-stated one be weighed rather than taken at face value, without ever putting words in the patient’s mouth.
Carrying the effort
The result is an exchange that meets a person where they are, in their own language and at their own pace, and hands the clinician a structured history they can trust, so the visit starts with understanding instead of an interrogation.
This is an evolving area of our work; we’ll publish methods and data as they mature.