From the doctor's plan to the patient's hand — and back
Remote care fails at the edges: the patient who never installs the app, the reading nobody chases, the alert that arrives too late. Here is exactly how CareSaathi AI closes each one.
Six steps, running continuously
Every patient in a programme is somewhere in this loop at all times. Nothing waits for a phone call.
Prescribe
The care team builds the plan: medicines and drops with times, exercises, education, and the monitoring programme — which metrics, how often, the target range, and the line at which someone must be told. AI drafts the first version from a description of the procedure; a clinician approves it before anything reaches a patient.
Reach
The plan goes to wherever the patient actually is. Connecting takes no technical skill: they message the care number and are recognised from the number already on their record, or staff connect it in one click, or they scan a QR. Nothing to download unless they want the app.
Guide
Reminders arrive at the right time, spoken aloud and shown with a photo of the actual bottle or strip — because “take your steroid drops” is a sentence, and a picture of the pink-capped bottle is an instruction. The patient confirms by saying “done”, tapping one big button, or tapping Done in WhatsApp.
Collect
Readings arrive four ways, and the platform does not care which: automatically from a connected device or wearable, by manual entry, from a photo of a monitor’s display that AI reads, or from a lab report the patient sends in. A dose that is never confirmed is recorded as missed — real adherence, not recall at follow-up.
Evaluate
Every reading is normalized to canonical units, de-duplicated, tagged with when it was observed and when it arrived, and checked against the doctor’s rules — thresholds, component rules, and persistence rules like “three consecutive abnormal readings”. Implausible values are stored as suspect rather than trusted blindly.
Act
Trends, adherence and de-duplicated alerts surface in one command center, ranked by who needs attention today. The team replies from the portal or straight from their own WhatsApp, and anything urgent escalates to a caregiver.
The same programme, three ways
The care team prescribes once. What changes is only how it reaches the patient — and the record they all write to is identical.
On WhatsApp
- The patient messages the care number — or staff connect the number already on their record.
- Reminders arrive with the medicine’s photo and Done / Skip buttons.
- They photograph a BP monitor or a lab report; AI reads the values into the chart.
- They ask a question and get an answer from their own approved plan — or a human.
- STOP pauses everything immediately; START resumes.
On their own phone
- The app pairs with a QR code or a short pairing code — no account to create.
- “Hello doctor” wakes a hands-free voice assistant grounded in their care plan.
- Full-screen spoken reminders; the patient says “done” to confirm.
- Exercise and education videos play in their language at the scheduled time.
- Everything works offline and syncs when a connection returns.
On a dedicated device
- Any tablet becomes a managed bedside device in seconds via QR pairing.
- A calm ambient screen: the time, the recovery day, the next reminder, one talk button.
- Kiosk mode locks it down for shared and ward use.
- Hospital branding, contacts and emergency numbers per facility.
- Full fleet visibility and remote management from the command center.
What actually happens on a Tuesday
One patient recovering at home, one care team, no phone calls.
A reminder arrives with the antibiotic drop’s photo. They tap Done. Adherence updates instantly.
They photograph their BP monitor. AI reads 128/82, files it as a preliminary reading and thanks them.
The reading is normalized, checked against the doctor’s thresholds, and matched to the expected morning slot.
“Is it normal that my eye waters?” — answered from the doctor’s approved recovery content, in their language.
The same question appears in the inbox with the AI’s answer attached, in case a human should add anything.
A nurse replies from her own WhatsApp. The patient receives it from the hospital’s number; it’s recorded in the chart.
An evening dose goes unconfirmed for two hours. It’s logged as missed, and after the third the caregiver is told.
Illustrative timeline of capabilities described elsewhere on this site.
Pre-op, post-op, chronic and elder care
Care items are scheduled relative to the procedure date, so a single plan spans the phases without anyone re-writing it.
Before a procedure
Day −14 to day 0Baseline vitals, pre-op tests, fasting instructions and which medicines to stop — each scheduled relative to the surgery date, so day −7 and day −1 arrive on their own. On surgery day the plan switches to post-op by itself, with no one flipping a status.
After a procedure
The fragile first weeksDrops and doses confirmed dose by dose, wound and report photos read by AI, daily check-ins for pain and symptoms, and warning-sign detection that routes anything worrying to a human immediately.
Ongoing & chronic care
Months to yearsHypertension, diabetes, heart failure and respiratory programmes running between visits — readings collected automatically, trends compared against the target band, and a climbing number surfaced as an alert rather than a readmission.
Elder & independent living
ContinuousBig type, spoken prompts and a caregiver loop. A patient who cannot use an app can still be monitored — on WhatsApp with one tap, or a bedside screen with one button — and a family member is told when something drifts.
Grounded by design, with a human always behind it
The AI is used where it is genuinely reliable — reading a number off a display, drafting a plan for a clinician to approve, answering from content a doctor already signed off — and stopped everywhere else.
- Answers come only from the clinician-approved plan, never the open internet
- It cannot change a dose, add a medicine, or give a diagnosis
- Anything it can't answer confidently goes to a human instead of a guess
- Urgent wording bypasses the assistant entirely and alerts the care team
- Values read from a photo are stored as preliminary until a clinician confirms them
- Every reading keeps its provenance: what read it, when, and from where