All use cases

Prevent / Care

Photo-based symptom triage and routing

A patient snaps a photo of an angry red patch around their injection site and asks "is this normal?" — the agent looks, decides how urgent it is, and either reassures or gets a clinician on it fast.

Next best thingClose the loopBehavioral triggerChatWhatsAppIn-appHealth & TelehealthWellness (B2C apps)

What it does

When something appears on the skin — a rash, a reaction at an injection site, a mole that has changed — a patient's first instinct is to photograph it and ask whether they should worry. Today that photo lands in a queue and waits, or it never gets sent and the concern festers; either way the patient is left guessing about something visible and frightening, and the genuinely urgent case waits in the same line as the harmless one. The agent receives the image on chat or WhatsApp, uses vision to assess apparent severity against the patient's history and program, and routes accordingly: clear, safe guidance and self-care signposting for the low-concern case, and an immediate clinician hand-off for anything that looks urgent — all logged to the record. It reads severity and routes; it does not diagnose or prescribe.

How it works

  1. 1
    Trigger. a patient sends a photo of a visible symptom in a conversation — an injection-site reaction, a rash, a skin concern — with or without a question attached.
  2. 2
    Decision. the agent runs vision over the image to gauge apparent severity, reads it against the patient's history and program stage (a known side effect of their medication, a recurring complaint, a brand-new lesion), and classifies it with conservative defaults — anything ambiguous or potentially serious escalates.
  3. 3
    Action. a low-concern result gets calm, approved guidance and self-care signposting plus what to watch for; a concerning result is routed straight to a clinician with the photo and history attached; the agent never names a diagnosis or recommends a treatment, and it asks for a clearer or additional photo when the image is inconclusive.
  4. 4
    Follow-through. the image, the severity read, and the routing decision are logged to the patient's record so a clinician opens it with full context; where guidance was given, the agent schedules a check-back to confirm the symptom settled, and any "it's worse" reply re-triages and escalates immediately.
Example
Patient[sends photo] is this normal after my shot? it's red and a bit warm
AgentThanks for the photo — some redness and warmth at the injection site in the first day or two is a recognized, usually minor reaction. For now you can keep an eye on it; please send me another photo if the redness spreads beyond the area, it gets more painful, or you feel feverish, and I'll get a clinician to look right away. I'll check back with you tomorrow to see how it's settling.

Configuration

How the agent is wired for this use case.

Triggeran inbound image in a patient conversation (photo_received) on chat, WhatsApp, or in-app, optionally with an accompanying symptom question.
Tools & actions
  • Vision model · assess the image for apparent severity and visible red-flag features, and request a clearer photo when the image is inconclusive.
  • Patient memory / history · read medication, program stage, and prior symptom reports to weigh the image in context.
  • Clinician escalation queue · hand off any concerning or ambiguous case to a clinician with the image and history attached.
  • Knowledge base · retrieve the program's approved self-care guidance and the watch-for / escalate-if criteria.
  • Care-record system · log the image, the severity read, and the routing decision for the clinical record and audit.
Autonomyimage intake, low-concern signposting from approved guidance, and the check-back schedule run unattended under judge gating. The agent gives NO diagnosis, dosing, or treatment advice — it reads apparent severity and routes only, under a clinical-safety policy class; any concerning, ambiguous, or potentially urgent image is a mandatory clinician hand-off.
Channelschat · whatsapp · in-app
Escalationa high- or uncertain-severity read, a red-flag visual feature, a worsening follow-up reply, or any request for a diagnosis or treatment decision routes immediately to a clinician with the image and patient context attached.

What you need

The inputs this use case runs on. Your channels stay yours; the agent supplies the judgment.

Signals

inbound image events in patient conversations, optional accompanying text, and patient follow-up replies.

Data

patient history, medication and program stage, prior symptom photos and reports, consent for image handling, persistent memory, the program's approved guidance and red-flag criteria.

Guardrails

the agent assesses severity and routes only — no diagnosis, dosing, or treatment advice ever; conservative defaults with mandatory clinician hand-off on any concerning or ambiguous image; explicit consent and secure, compliant handling of medical images and personal data; judge gating on every message; a clear scope boundary stated to the patient; full audit trail of every triage decision.

Metrics it moves

  • time-to-resolutiondown: a visual concern is triaged the moment the photo arrives, so the urgent case reaches a clinician immediately instead of waiting in a shared queue.
  • safe-escalation-rateup: concerning images are reliably surfaced to clinical eyes with conservative defaults rather than sitting unread or unsent.
  • contact-ratedown: clear, safe guidance on genuinely low-concern photos resolves the "is this normal?" question without consuming a clinical slot.

Related use cases

Prevent / Care

Clinical red-flag detection and safe escalation

the text-based symptom rail this extends to visual inputs

Engage / Lifecycle

Test-result interpretation agent

the same clinical-safety posture applied to lab results rather than images

Engage / Lifecycle

Proactive health-coach companion

the program companion the safe-guidance replies sit inside

See it on your own customer journey

Bring one drop-off, one churn cliff, or one silent segment. We will show you what a proactive agent with memory and judgment does with it.

Book a demo