Skip to content
KKoderClub

AI & Automation · Healthcare

An AI triage assistant cut patient intake time by 42% across nine hospitals

A nine-hospital network was losing outpatient capacity to manual intake. We built a clinician-supervised AI triage assistant that structures symptoms, routes to the right department and pre-fills the EMR — cutting intake time by 42%.

Client
Confidential hospital network
Region
India (multi-city)
Timeline
4 months to production
Team
6 specialists

42%

Faster average patient intake

9

Hospitals live on the assistant

96%

Clinician acceptance of routing suggestions

100%

Suggestions with an audit trail

Context

The situation we found

Nine-hospital private network with 2,400 daily outpatient visits

Engagement model: Discovery sprint + build + managed AI operations

Front-desk staff captured symptoms free-text, then a nurse re-interviewed the patient before routing. The duplication added an average of eleven minutes per patient and produced inconsistent department routing.

Challenge

What was blocking the business

The constraints that had to be resolved before any technology decision mattered.

Duplicate intake interviews between reception and nursing staff
Inconsistent department routing causing avoidable specialist referrals
Patient data handling requiring strict consent and audit controls
Multilingual patient base across Hindi, Gujarati and English
Zero tolerance for autonomous clinical decisions

Approach

How we solved it

The decisions that produced the outcome, not a feature list.

Human-in-the-loop by design

The assistant proposes; a clinician confirms. Every suggestion is reviewable, overridable and logged against the reviewing user.

Structured symptom capture

An LLM converts multilingual free-text and voice input into a structured intake record mapped to the network's own triage taxonomy.

Retrieval grounded in network protocols

Routing recommendations are grounded in the network's approved clinical protocols, not general model knowledge, with citations shown to the clinician.

Privacy and audit controls

Consent capture, PII redaction before model calls, regional data residency and a complete, immutable audit log.

Delivery roadmap

How the programme ran

  1. 01

    Discovery & risk assessment · 3 weeks

    Process shadowing, data protection review and success metric definition.

  2. 02

    Pilot build · 6 weeks

    Intake assistant, protocol retrieval and clinician review console for one hospital.

  3. 03

    Clinical validation · 4 weeks

    Supervised pilot, accuracy benchmarking and protocol tuning.

  4. 04

    Network rollout · 5 weeks

    Staged rollout across nine sites with training and managed AI operations.

Results

What changed after go-live

Measured in the client's own systems against the baseline captured in discovery.

Average intake time fell from 26 to 15 minutes per outpatient
Department mis-routing dropped sharply, reducing avoidable specialist referrals
Nursing staff redeployed from data entry to patient-facing care
Every AI suggestion is traceable to its source protocol and reviewing clinician

Client view

In their words

The assistant never makes a clinical call on its own — it removes the paperwork around one. That distinction is why our clinicians trust it.

Chief Medical Information Officer

CMIO, Confidential hospital network

Stack & integrations

What it was built on

Technology stack

PythonFastAPIReactPostgreSQLDocker

Integrations

Hospital EMRSMS/WhatsApp notificationsIdentity provider (SSO)Speech-to-text

FAQ

Questions about this engagement

Healthcare

Facing something similar in healthcare?

Share your current systems and constraints. We will return an approach, phased plan and the outcome metrics we would commit to — under NDA.

Request a proposal for healthcare

Share your current systems and constraints. A senior consultant responds within one business day, under NDA.

NDA friendly. We never share your details.

CallBook a consultation