Curve Metrics logo - web development company in Nagpur

Hospital AI & Emergency Healthcare Case Study

Nephron Kidney Hospital — Autonomous AI Receptionist & 24/7 Clinical Emergency Triage Agent

Production-grade autonomous AI receptionist for Dr. Jitesh Jeswani's Nephron Kidney Hospital in Nagpur. Features Claude 3.5 Sonnet agentic tool-calling, 24/7 emergency renal triage guardrails, bilingual Hindi/English voice AI, and a zero-code self-learning review loop.

Nephron Kidney Hospital 24/7 AI Receptionist Kiosk and Emergency Triage by Curve Metrics

Challenge

The business problem

Hospital front-desk operations face a dangerous paradox: routine administrative inquiries (consultation hours, OPD fees, doctor schedules, insurance tie-ups) overwhelm receptionist phone lines, causing delayed triage during critical medical emergencies. At Nephron Kidney Hospital—a premier nephrology, dialysis, and kidney transplant center in Nagpur led by Dr. Jitesh Jeswani—receptionists handled 180+ daily phone calls while managing crowded physical waiting rooms. At night (10 PM to 7 AM), anxious patients experiencing acute renal symptoms (acute breathlessness, sudden fluid retention, missed dialysis) had to navigate generic answering services. Furthermore, rigid scripted rule-based chatbots failed completely because clinical patient inquiries are nuanced, while raw unconstrained LLMs risked medical hallucination or prescribing medications. The hospital needed a deterministic, safe, production-grade agentic system capable of live scheduling without double-bookings and instant 24/7 emergency escalation.

"The emergency renal guardrails are what set Curve Metrics' AI apart. When a dialysis patient contacts us late at night feeling breathless, the AI doesn't give generic advice or schedule a routine visit for tomorrow—it instantly flags pulmonary edema risk, dispatches our 24/7 emergency dialysis desk, and directs them to the emergency ward. It has halved our routine front-desk phone load while giving critical patients a sub-2-second lifeline."

Solution

What Curve Metrics shipped

  • Autonomous Agentic Tool Calling (Claude 3.5 Sonnet): Replaced brittle decision trees with an autonomous reasoning agent executing typed tools (check_availability to validate OPD hours/doctor leaves in real time, and create_appointment to validate patient records with 0% slot clash).
  • Life-Critical Emergency Guardrails: Engineered strict deterministic safety guardrails. When clinical keywords like acute breathlessness, sudden severe edema/fluid overload, chest pain, or missed urgent hemodialysis are detected, the AI immediately overrides routine scheduling and displays an Emergency Quick Card with 1-tap dial to the 24/7 Dialysis Desk (+91 89567 10030) and live hospital navigation.
  • Bilingual Voice AI (Web Speech API): Integrated native client-side Speech-to-Text (STT) and Text-to-Speech (TTS) in English and Hindi. Enables elderly patients and rural caregivers across Vidarbha to speak naturally on their smartphones or hospital kiosk with zero audio API latency and zero per-minute audio charges.
  • Closed-Loop Self-Learning Architecture: Whenever the AI encounters ambiguous patient queries below a 98% confidence threshold, it gracefully defers and queues the interaction to an admin review portal. Hospital administrative staff can review and answer in one click, expanding the knowledge base instantly without modifying code.
  • Acid-Safe Pure-JS Transactional Engine on Shared Hosting: To operate reliably on existing hospital infrastructure without heavy native C++ database dependencies that trigger NPROC limits, engineered an ACID-compliant memory-mapped transactional store delivering sub-millisecond retrieval.

Results

Measured outcomes

< 2s Instant 24/7 Patient Triage & Emergency Response

52% Reduction in Routine Front-Desk Inbound Phone Calls

0% Double-Bookings or OPD Scheduling Clashes Over 6+ Months

3.4x Increase in Pre-OPD Digital Patient Intake Registrations

Planned operations CRM

From PO entry to dispatch and management review

The planned sequence is grouped below for readability. These are roadmap steps, not a list of features already delivered. The final manager step and acceptance rules will be clarified during discovery.

  1. Step 1: Patient or attendant opens hospital portal or physical kiosk, initiating voice or text interaction in Hindi or English.
  2. Step 2: Emergency Symptom Classifier runs deterministically. If life-critical renal indicators trigger, emergency card with direct 24/7 Dialysis Desk hotline (+91 89567 10030) fires in < 500ms.
  3. Step 3: For routine OPD inquiries, Claude 3.5 Sonnet agent invokes check_availability tool against live doctor schedules and leaves.
  4. Step 4: AI captures patient name, mobile, age, and chief complaint, invoking create_appointment to lock slot and generate digital appointment pass.
  5. Step 5: Automated WhatsApp confirmation message with location link, doctor instructions, and Google Calendar invite sent to patient's phone.
  6. Step 6: Unknown or low-confidence questions auto-route to staff review queue, continuously training the system without code redeployment.

Proposed validation

How we plan to measure the work

  • Audit logs of 1,200+ patient interactions verified zero medical advice or prescription hallucinations.
  • Front-desk call duration metrics before vs after AI deployment recorded across 90 operating days.
  • Real-time emergency escalation latency measured on mobile 4G/5G connections (median response 1.4 seconds).
  • Live verification available directly on production site: https://nephronkidneyhospital.com
Explore the FIBC manufacturing CRM planning guide
Interactive Workflow Simulator

See How Social & Email Automation Transforms Operations

Dental Desk 5-Step Patient Nurture Sequence:

42% Average Email Open Rate
1

Trigger: Instant

Website Lead Capture

Patient fills out consultation form or clicks WhatsApp button.

Syncs to CRM
2

Time: +1 Minute

Welcome & Brochure Email

Sends personalized treatment guide + doctor credentials via email.

64% Open Rate
3

Time: +24 Hours

WhatsApp Booking Link

Automated WhatsApp ping with 1-tap OPD slot selector.

88% Read Rate
4

Time: Day 4

Patient Case Study Drip

Sends before/after care tips & patient video testimonials.

Trust Signal
5

Outcome

Confirmed Appointment

Patient confirms slot; front-desk receives automated alert.

3.8x Conversion

Zero Manual Effort: Once configured, every incoming lead receives personalized care guides and follow-ups automatically 24/7.

Read Dental Desk Case Study

Start a project

Want a result-focused digital strategy?

Get a clear roadmap for your website, ecommerce store, AI agent or SEO lead funnel.