Product Requirements Document

Clinic Saathi

An agentic AI assistant that runs a clinic's operations entirely on WhatsApp — the app the doctor, the staff and the patients already use. It replaces the booking website, the patient app and the practice-management CRM with one conversation.

Version 1.0 · DraftDate June 2026Status For reviewClassification Confidential

1. Overview

1.1 Purpose

This document defines the product requirements for Clinic Saathi, an agentic system for small and mid-size clinics. Instead of a booking website, a patient app and a practice CRM, the clinic gets one AI agent that lives on WhatsApp and runs five areas of the practice: Front Desk (booking, reminders, no-show recovery, intake), Billing (payment links, invoices, dues), Staff (attendance, leave, shifts), Inventory (stock & expiry alerts, re-order) and Reports (ask a question, get an answer in seconds).

1.2 Product vision — replace the CRM, the app and the website

The goal is to remove software, not add it. Small clinics don't adopt dashboards, patient apps or CRMs — staff turnover is high, training is hard, and there's no budget. WhatsApp is the one interface everyone already knows. Clinic Saathi turns that familiar chat into the entire operating system of the clinic: patients book, confirm, pay and fill intake in chat; staff check in and request leave in chat; the doctor asks "how much did I make last week?" and gets the answer in chat. No app to install, no dashboard to learn, no CRM to maintain.

1.3 What stays off WhatsApp

A few things genuinely can't (and shouldn't) be replaced by a chat message — so the agent hands them off with a single tap and brings the result back into the conversation: the payment checkout (a secure Razorpay link), the Google review page (a one-tap link), and documents like the PDF invoice and the weekly revenue report (delivered as WhatsApp document messages, forwardable to an accountant). Everything else — the decisions, the prompts, the confirmations — happens in the thread.

1.4 Scope

In scope: a WhatsApp-Business-API agent covering appointment booking, reminders & no-show recovery, pre-visit intake, payment links & collection, invoices & dues, staff attendance & leave, inventory & expiry alerts, on-demand and scheduled reports, and review prompts — delivered in two phases (V1, V2).

Out of scope (v1): EHR / clinical records and prescriptions, insurance/TPA claims, tele-consultation video, multi-clinic franchise management, and any native mobile app or web dashboard (deliberately — the product's premise is that none is needed).

2. Personas

Everyone interacts through the same WhatsApp number; the agent recognises who is messaging and adapts.

PersonaDescriptionWhat they do on WhatsApp
PatientBooks and visits the clinic.Book/confirm/reschedule, fill intake, pay, get invoice, leave a review.
Doctor / Clinic ownerRuns the practice.Ask reports, get a morning brief, approve invoices & leave, see alerts.
Front-desk / ReceptionistManages the day's flow.See the schedule, handle walk-ins, mark dues, update stock.
Clinic staffNurses, assistants, pharmacist.Check in/out, request leave, update inventory in plain language.
Supplier V2Medicine distributor.Receives auto re-order messages and confirms.

3. Front Desk Agent BUILD FIRST

3.1 WhatsApp appointment booking V1

A patient messages any time; the agent checks live availability and confirms a slot with doctor name, date and time. Interactive WhatsApp buttons let the patient pick a slot without typing — the conversation does the work a booking website would, with zero learning curve. See the booking flow play out in the demo →

3.2 Smart reminders + no-show recovery V1

Automated 24-hour and 2-hour reminders with tap-to-confirm. If a patient doesn't respond, the agent auto-triggers a waitlisted patient to fill the slot. Indian clinics using structured WhatsApp reminder flows report up to a 78% reduction in no-shows in documented case studies.

3.3 Pre-visit intake form V1

A 4-question WhatsApp intake before the visit; the doctor receives a 3-line AI summary instead of a blank chart — saving 3–4 minutes per consultation. The intake is conversational, not a form to fill on a website.

3.4 Missed-call recovery V2

A missed call to the clinic triggers an instant WhatsApp: "We missed your call — tap to book," converting an otherwise-lost lead. Indian telephony providers (Exotel, Knowlarity) offer missed-call-to-webhook triggers built for exactly this.

4. Billing & Payments

4.1 Payment link + UPI collection START HERE

After each visit the agent sends a Razorpay payment link — UPI, card or net-banking — enabling same-day collection instead of a 3–7 day delay. Razorpay now ships Agentic Payments designed for AI agents to transact within consent-based, pre-authorised limits — built for exactly this use case.

4.2 Auto invoice generation V2

On visit completion the agent generates an itemised PDF invoice and sends it as a WhatsApp document; the doctor approves with one tap. The document leaves the chat only as a file the patient (and accountant) can keep.

4.3 Outstanding dues tracker V2

The agent monitors unpaid invoices, sends reminders at 3 / 7 / 14 days, and flags habitual late payers — mirroring Razorpay's own "Subscription Recovery" agent pattern.

5. Staff Scheduling & Attendance

5.1 WhatsApp attendance check-in START HERE

Staff text "in" / "out" at shift start and end; the agent logs the timestamp, flags late arrivals and sends the owner a daily summary. Zero hardware — it replaces the biometric machines small clinics can't justify.

5.2 Leave request + approval flow V2

Staff request leave via WhatsApp; the agent checks coverage and auto-approves if minimum staffing holds, otherwise routes to the doctor for a one-tap approval.

6. Pharmacy & Inventory

6.1 Low-stock + expiry alerts START HERE

A daily check against reorder thresholds and expiry dates sends a WhatsApp alert 30 days before expiry. This is a pure rule engine — no AI required — which makes it cheap and fast to ship, and a great early win.

6.2 Medicine inventory tracking V2

Staff update stock in natural language ("paracetamol 500mg add 200 strips"); the agent parses and records it. No barcode scanner needed — it works on a ₹5,000 Android tablet.

7. Reports & Analytics

7.1 On-demand query — "ask your clinic" BEST DEMO · BUILD 2ND

The doctor types "How much did I earn last week?" — the agent runs text-to-SQL against the clinic's database and replies in plain language in seconds. No dashboard, no login, no learning curve. Once the database has data, this takes roughly two days to build.

7.2 Daily morning brief V2

An 8 AM WhatsApp message: today's appointments, yesterday's revenue, pending payments, staff present, and one key alert — so the doctor is informed before walking in. No login required.

7.3 Weekly & monthly revenue report V2

An auto-generated PDF every Monday and on the 1st — revenue breakdown, top patients, no-show trend and top medicines — forwardable straight to the clinic's accountant.

8. Reputation & Reviews V2

After a positive interaction — for example, a follow-up confirming the patient is feeling better — the agent prompts satisfied patients to leave a Google review via a one-tap WhatsApp link. Review volume directly drives new-patient discovery for clinics with no marketing budget, and this is a natural, low-effort extension of the existing post-visit follow-up: same trigger, one more message.

9. Non-Functional & Compliance

WhatsApp policy

Built on the WhatsApp Business Platform; respects 24-hour session windows, approved message templates and opt-in consent.

Privacy & health data

Encrypted PII, consent capture, data-minimised intake, role-based access; aligned with India's DPDP Act.

Performance

Agent replies in under 3 seconds; reminders and briefs fire on schedule within the minute.

Reliability

Idempotent webhooks & payments, retry queues, and graceful fallback to a human handoff when unsure.

AI guardrails

Text-to-SQL is read-only and scoped to the clinic's data; the agent never invents figures and cites the period it queried.

Observability

Full conversation & action audit log, payment reconciliation, and alerts on failed sends or webhooks.

10. Build Phases (V1 / V2)

PhaseFeaturesWhy this order
V1 — CoreWhatsApp booking, reminders + no-show recovery, pre-visit intake, payment link, attendance check-in, low-stock/expiry alerts, on-demand "ask your clinic" query.Highest pain, fastest payback; payment link + attendance + alerts are quick wins, booking is the anchor, reports are the best demo.
V2 — ExpandAuto invoices, dues tracker, leave flow, natural-language inventory, daily morning brief, weekly/monthly PDF reports, missed-call recovery, review prompts.Build on the V1 data and post-visit triggers already in place — each is incremental, not net-new plumbing.

11. Open Questions

  1. WhatsApp access: official Cloud API direct, or via a BSP (Gupshup / AiSensy / Wati) for templates & number provisioning?
  2. One shared clinic number for all personas, or separate numbers for patients vs staff/owner?
  3. Which LLM provider for intake summaries & text-to-SQL, and on-prem vs API given health-data sensitivity?
  4. Pricing model to the clinic — per-clinic annual platform fee plus WhatsApp/template & AI usage at actuals?
  5. Which V1 features go live in the first 2-week pilot for the launch clinic?
Interactive demo

It all happens in the chat.

No app, no dashboard, no CRM — just WhatsApp. Pick a side (patient or clinic owner), tap a scenario, and watch the agent work. The quick-reply buttons are live — tap them.

Delivery timeline

A proposed 12-week build.

Five phases. V1 ships the highest-pain, fastest-payback features first; V2 builds on the same data and post-visit triggers.

Front Desk Billing & Reports Staff & Inventory AI / Agent Milestone

Phase 1 · Foundation & Front Desk

WhatsApp Business API & webhooks, booking with live availability & buttons, reminders + no-show recovery, pre-visit intake with AI summary.

Weeks 5–7

Phase 2 · Billing & Reports

Razorpay payment links & collection, and the "ask your clinic" text-to-SQL query — the best demo.

Weeks 8–9

Phase 3 · Staff & Inventory

WhatsApp attendance check-in & daily summary, low-stock & expiry alerts (rule engine).

Phase 4 · V2 Expansion

Invoices & dues, leave flow, natural-language inventory, daily brief, weekly/monthly PDFs, missed-call recovery, reviews.

Week 12

Phase 5 · Hardening & Handoff

WhatsApp policy & template review, security & privacy hardening, reconciliation, UAT with the pilot clinic & documentation.

Technology

One interface. One agent. No dashboard.

WhatsApp is the only surface the clinic ever sees. Behind it, a NestJS agent backend orchestrates booking, billing, staff, inventory and reports, with an LLM layer for intake summaries and natural-language reporting.

Interface

WhatsApp Business Platform

The whole product. A verified business number using the Cloud API (or a BSP) — interactive buttons & lists, templates, and document messages.

  • Buttons, lists & quick replies
  • Approved message templates
  • Document messages (invoices, reports)
  • Opt-in & 24h-window handling
Agent backend

NestJS

A modular agent that handles conversation, booking, billing, staff, inventory and reports, plus a scheduler for reminders and briefs.

  • Node + TypeScript, modular services
  • Webhook ingestion & routing
  • Redis job queue for reminders/cron
  • Idempotent payments & retries
Intelligence

LLM layer

Language understanding for intake summaries and natural-language inventory, and guard-railed text-to-SQL for "ask your clinic".

  • Intent & entity parsing
  • 3-line intake summaries
  • Read-only text-to-SQL with guardrails
  • Human handoff when unsure

System architecture

PatientsWhatsApp
StaffWhatsApp
Doctor / OwnerWhatsApp
messages
WhatsApp Cloud API / BSPtemplates · buttons · documents
Webhook Gatewayverify · route · session
NestJS agent services
Conversation & NLU
Booking & Reminders
Billing & Payments
Staff & Attendance
Inventory
Reports · text-to-SQL
Scheduler & Jobs
AI / LLM gateway
PostgreSQLClinic data
RedisJobs · cron · session
Object storageInvoices & reports
Outbound
LLM providersummaries · text-to-SQL
Razorpaylinks · agentic payments
Exotelmissed-call → webhook

Supporting stack

Data

PostgreSQL (clinic data), Redis for the reminder/brief job queue & session state.

Payments

Razorpay payment links + Agentic Payments, webhooks, reconciliation & auto-refunds.

AI

LLM for intake summaries & NL inventory; read-only, guard-railed text-to-SQL for reports.

Telephony

Exotel / Knowlarity missed-call-to-webhook for lead recovery.

Infra & DevOps

Containerized (Docker) on AWS/GCP, CI/CD, autoscaling workers for reminders.

Observability

Conversation & action audit logs, payment reconciliation, alerts on failed sends.

Why this approach

Zero adoption cost

Nothing to install or learn. Staff and patients use WhatsApp already — the #1 reason clinic software fails is adoption, and this removes it.

Replaces three tools

One agent does the job of a booking website, a patient app and a practice CRM — fewer logins, fewer subscriptions, one source of truth.

Fast & incremental

Rule-engine wins (alerts, attendance) ship in days; the agent & reports build on the same data without net-new plumbing.

Built for India

Razorpay agentic payments, Exotel missed-call, WhatsApp-first behaviour and ₹5,000-tablet operability — designed for small clinics here.