P3 · Clinically adjacent — highest risk
Bolo Follow-Up
Post-visit check-ins that answer from your own instructions.
Post-op day 3 · Hindi + English
Not clinically reviewedIt never assesses a symptom, never gives a dose the discharge instructions don't contain, and never answers “should I go to the ER?” with anything but escalation. These are hard-coded refusals, not prompt instructions.
- Patient · 00:18 · Hindi + English
- dressing कब बदलना है? When should I change the dressing?
- Patient · 00:51 · Hindi
- छाती में भारी लग रहा है. There's a heaviness in my chest.
- ANSWERABLE — Source: discharge instructions, v3, signed
- Read back from this patient's own discharge instructions, in Hindi.
- RED FLAG — Tier 3 · escalated
- Escalation script → on-call provider paged → stayed on the line
Synthetic demo call. Not a real patient.
POD 1·3·7·30
the cadence a protocol asks for, and the one that slips first
Post-discharge call volume lands on the nurses, and the calls that most need making are the ones hardest to make — a post-op patient who speaks Hindi, at home, on day three, with a question about a dressing. Those calls get shortened or skipped, and the readmission is the consequence.
How it works
-
01
Call at a time the patient chose, in the language they chose.
-
02
Classify every patient turn three ways: answerable, unknown, or red flag.
-
03
Answer only from this patient's discharge instructions and your clinician-signed protocol library.
-
04
Escalate anything else up a logged, timestamped ladder that ends with a human on the line.
The escalation ladder
Every turn is classified before anything is answered.
The classifier is tuned for recall on red flags, not precision. A false escalation costs a nurse two minutes; a missed one costs a life and the company. Your nurses will field more escalations than they strictly need to, and we would rather tell you that now than have you find it out in week three.
-
Answerable → Tier 0
Retrieved from this patient's discharge instructions or your version-signed protocol library. Nothing else is in scope — not the open web, not general medical knowledge, not model priors.
-
Unknown → Tier 1 · 2
Anything the system cannot ground in a signed source. It says so plainly and hands off rather than reaching for an answer.
-
Red flag → Tier 3 · 4
Fired by a colloquial-register lexicon built per language by linguists, not translated from a list of clinical terms. This has to fire as reliably as “chest pain”:
छाती में भारी लग रहा है
-
Tier 0 Bolo answersAnswerable, grounded in a signed source Immediate
-
Tier 1 Nurse queue, asyncUnknown 4 business hours
-
Tier 2 Nurse callbackUnknown, with distress detected 60 minutes
-
Tier 3 On-call provider pagedRed flag Immediate
-
Tier 4 Emergency script, stays on the lineCritical red flag Immediate
Every tier transition is logged, timestamped and reportable. The nurse queue is a product surface with the transcript, the detected flag and one-click callback — not an email alias.
What it will not do
Stated plainly, because a limit you find out about later is a limit you stop trusting the rest of the page over.
- It is not a nurse and we do not call it one. Nursing is a licensed practice and the title is protected.
- It never assesses a symptom. It never says “that sounds normal” or “you're fine”.
- It never gives a dose the discharge instructions do not contain, and never adjusts a medication.
- It never answers “should I go to the ER?” with anything other than escalation. These are hard-coded refusals, not prompt instructions.
- It answers from two sources only: this patient's instructions, and your version-signed protocol library. Not the open web, not general medical knowledge, not model priors.
Integration
Every buyer asks within ninety seconds of the demo starting. A plain FHIR path exists today — “here is the bundle” is a real answer, and “we are in the marketplace review process” is not.
- FHIR Task
- Nurse queue items, with the transcript and the detected flag attached.
- Protocol library
- Version-signed by a named clinician. The system will not answer from an unsigned version.
- SMS fallback
- In the patient's language, when a call is missed. Never abandoned silently.
- Outbound telephony
- Prior express consent captured at intake, revocation honoured immediately.
What we measure
These are the numbers for this product. None of them are published yet — see the methodology for why an unstratified figure would be worse than silence.
- Red-flag recall
- Must be at or near 100%. We report the misses.
- Escalation precision
- Expected to be low, deliberately. A false escalation costs a nurse two minutes.
- Answer groundedness
- Percentage of answers traceable to a signed source document. Target 100%.
- Time to nurse callback
- Per escalation tier, against the stated SLA.
What you are probably about to ask
Is this an AI nurse?
No, and we will not call it one — nursing is a licensed practice and the title is protected. It delivers your own approved discharge instructions in the patient's language and routes everything else to your nurse. It does not assess, advise or triage.
What stops it from inventing medical advice?
It can only answer from two sources: this patient's own discharge instructions, and your protocol library, version-signed by a named clinician. Not the open web, not general medical knowledge, not model priors. A question it cannot ground goes to the nurse queue.
How many false escalations will our nurses field?
More than you would like, on purpose. The classifier is tuned for recall on red flags, not precision. A false escalation costs a nurse two minutes; a missed one costs a life and the company. We would rather tell you that up front than have you discover the tuning later.
Pilot
Book a pilot call.
Thirty minutes. We will run the demo on the languages your panel actually speaks, show you the eval methodology, and tell you plainly which parts are not ready. If we are not a fit we would rather find out on this call.
One reply from a person. We do not run a nurture sequence and we do not share your address.