2 ready-to-build workflows

AI agent workflows for Hospitals & health systems

Automate OPD scheduling, discharge follow-up and multilingual patient helpdesk — with clinicians deciding anything clinical.

Each recipe below is expressed only in VegaDūta's real workflow building blocks — Trigger, Agent, Knowledge, Condition, Approval, Output, Tool (MCP), Voice, Device, Loop, Code and Parallel — so it maps 1:1 to something you can assemble in the Workflow designer. Consequential actions always pass a human Approval step.

1. OPD appointment booking with pre-visit instructions

Patients call the front desk to book an OPD slot, and staff manually check the doctor's calendar, book the visit, then read out fasting/document instructions — a queue that only moves in office hours and only in one language.

Trigger: WhatsApp / Call (patient booking request)

How the workflow runs

Trigger
Patient requests a slot. A WhatsApp or Call trigger starts on any inbound booking request, in whatever language the patient speaks.
Agent
Understand the need + find a slot. An Agent node captures the department, doctor and preferred time, working in the patient's language via Sarvam.
Tool (MCP)
Book in the HIS. A Tool (MCP) node reserves the slot in your HIS / scheduling system. There's no first-party connector — the booking is pushed via MCP (or queued to the front desk).
Knowledge
Attach the right prep instructions. A Knowledge node pulls the correct pre-visit instructions (fasting, documents, arrival time) from your hospital's own protocol documents — grounded, with citations, never invented.
Output
Confirm + remind. An Output node sends the confirmed slot and prep checklist on WhatsApp/SMS, then a day-before reminder to cut no-shows.

Channels & connectors

WhatsApp
Voice / Call
Sarvam (22+ Indian languages)
Knowledge base
SMS
Tool (MCP → HIS/scheduler)

Outcome

Patients book an OPD slot and receive the exact pre-visit prep in their own language, day or night, without tying up the front desk.

Why it helps

Replaces the manual calendar-check and read-out-the-instructions call with an always-on, multilingual booking flow — grounded prep comes from your own protocols, not a guess.

Build spec

1 agent — 1 booking agent; the HIS booking, the prep-instruction Knowledge pull and the reminders are workflow nodes.

System prompt (paste-ready)

You are an OPD booking assistant for {{hospital_name}}. Working in the patient's language, capture the department, doctor and preferred time and confirm the details. Offer only slots the scheduling system returns as real. Attach pre-visit instructions ONLY from the hospital's own protocol documents in Knowledge, with citations. You do NOT give medical advice, interpret symptoms, triage urgency, or suggest which doctor is clinically right — if the patient describes an emergency, tell them to call emergency services or go to the ER. Don't collect clinical history beyond what booking needs.

MCP connectors

HIS / scheduling system — via a Tool (MCP) node (no first-party connector)
WhatsApp + Telephony — via the WhatsApp/Voice channels
Sarvam — 22+ Indian languages

Built-in tools

knowledge_search (pre-visit protocol documents)
http_request (check live slot availability)

Guardrails

No clinical advice, symptom interpretation, or urgency triage — an emergency is redirected to emergency services, never assessed by the agent
Prep instructions come only from the hospital's own protocol documents with a citation — never invented
Only books slots the HIS confirms as available (anything it can't book is queued to the front desk), and only booking-necessary details are collected

Output & delivery

The agent captures the request in the patient's language → a Tool node reserves the slot in the HIS via MCP (or queues it to the front desk) → a Knowledge node attaches the grounded prep checklist → an Output node sends the confirmation and a day-before reminder on WhatsApp/SMS.

2. Discharge follow-up & medication-adherence check-ins

After discharge, a nurse is supposed to call each patient to check they're taking medication and recovering — but at ward volume most calls never happen, and warning signs are missed until a readmission.

Trigger: Webhook (discharge event) — scheduled check-ins

How the workflow runs

Trigger
Patient is discharged. A Webhook trigger (from your HIS/EHR) opens a follow-up plan the moment discharge is recorded — no first-party connector, the event arrives via Webhook.
Loop
Scheduled check-ins over recovery window. A Loop node paces check-ins across the recovery period (e.g. day 2, day 7, day 14) rather than a single message.
Agent
Ask about meds + recovery. An Agent node checks medication adherence and recovery against the discharge plan, in the patient's language — it asks and listens, it does not diagnose.
Condition
Flag anything concerning. A Condition node detects reported red-flag symptoms, missed doses or distress and branches those cases out of the routine track.
Approval
Clinician reviews flagged cases. An Approval node routes any clinical concern to a nurse or doctor for the judgement call — the agent never assesses or advises on the condition itself.
Output
Log + reassure routine cases. An Output node records the check-in against the patient and sends routine adherence reminders on WhatsApp/SMS.

Channels & connectors

Webhook
WhatsApp
SMS
Sarvam (multilingual)
Approval / HITL
Tool (MCP → HIS/EHR)

Outcome

Every discharged patient gets consistent, in-language check-ins and any warning sign reaches a clinician fast — instead of follow-up calls that skip most of the ward.

Why it helps

Removes the manual per-patient follow-up that doesn't scale, while every clinical judgement stays with a licensed human — the agent surfaces, the clinician decides.

Build spec

1 agent — 1 check-in agent; the paced Loop, the red-flag Condition and the clinician Approval are workflow nodes.

System prompt (paste-ready)

You are a post-discharge check-in assistant. In the patient's language, ask about medication adherence and recovery against their discharge plan, and record their answers as {doses_taken, symptoms_reported[], concerns, needs_clinician: boolean}. You ASK and LISTEN — you do NOT diagnose, interpret symptoms, change a dose, or advise on the condition. Set needs_clinician=true and stop the routine track on any reported red-flag symptom, missed-dose pattern, or distress. If the patient describes an emergency, tell them to call emergency services or return to hospital immediately.

MCP connectors

HIS / EHR — discharge event inbound via Webhook, records written via a Tool (MCP) node (no first-party connector)
WhatsApp + SMS — via the messaging channels
Sarvam — multilingual

Built-in tools

knowledge_search (discharge plan + red-flag criteria)
transfer_to_human

Guardrails

Never diagnoses, advises on the condition, or changes medication — it asks, listens, and records; every clinical judgement is the clinician's at the Approval node
Any red-flag symptom, missed-dose pattern, or distress branches out of the routine track to a clinician; an emergency is redirected to emergency services
Check-ins are logged against the patient record; clinical detail is handled as confidential and not shared beyond the care flow

Output & delivery

A Loop node paces check-ins across the recovery window → the agent asks and records adherence/recovery in the patient's language → the Condition node flags anything concerning → a clinician reviews flagged cases at the Approval node, while routine cases are logged and get adherence reminders on WhatsApp/SMS.

Related industries

Build one of these in minutes

The sandbox gives you a live agent workspace — no account, no card. Or head back to the full catalogue and compare patterns across every industry.

Try this in the sandbox — no signup Browse all industries

See how VegaDūta compares to n8n, Dify and BotpressWhy VegaDūta's architectureEstimate your WhatsApp API costs