Table of Contents

Overview

Problem

Key Challenges

The Approach

Timeline

Next Steps

Key Learnings

generative ai • live 2026

Creating a Generative EMR

Redefining manual form filling for EMRs

Role

Product Designer & Solution Architect

Timeline

3 Months

Team

2 Product Designers & 3 Product Mangers

overview

What if we could transform a traditional EMR into an intelligence-assisted clinical documentation system?

Clinicians spend up to half their workday on EHR and desk work often more than they spend with patients. The Enhanced EMR replaces manual form-filling with conversational AI: doctors talk or type, and the system generates structured medical records in real time.

Clinicians spend up to half their workday on EHR and desk work often more than they spend with patients. The Enhanced EMR replaces manual form-filling with conversational AI: doctors talk or type, and the system generates structured medical records in real time.

Existing EMR at Bajaj Finserv Health

The Enhanced EMR reimagines this workflow with conversational AI and generative intelligence, enabling clinicians to interact with the system via natural language (typed or spoken) while the system automatically interprets and generates structured medical records in real time.

The Enhanced EMR reimagines this workflow with conversational AI and generative intelligence, enabling clinicians to interact with the system via natural language (typed or spoken) while the system automatically interprets and generates structured medical records in real time.

Problem

Clinical Documentation Today

  1. Physicians spend ~36 minutes per patient on EHR tasks, often after hours.

  1. Physicians spend ~36 minutes per patient on EHR tasks, often after hours.

  1. For every hour of direct patient care, clinicians can spend two additional hours on clerical work.

  1. For every hour of direct patient care, clinicians can spend two additional hours on clerical work.

  1. Many physicians report documentation as a core driver of burnout and reduced job satisfaction.

  1. Many physicians report documentation as a core driver of burnout and reduced job satisfaction.

How a Clinician’s Time Is Actually Spent

45% EMR

30% patient

25% admin

EMR at Bajaj Finserv Health

Usage was growing fast which raised the stakes for getting this right.

Usage was growing fast which raised the stakes for getting this right.

7,16,280

Prescriptions Generated as compared to 2,21,728 in 2022

223% from the old version

15,798

Active Doctors as compared to 7,160 in 2022

75% from the old version

Clearly this brings a need to enhance and introduce new features

Clearly this brings a need to enhance and introduce new features

key challenges

The goal was not to redesign the system, but to evolve it without disruption.

Every enhancement had to balance technical, behavioral, and philosophical constraints at once.

Every enhancement had to balance technical, behavioral, and philosophical constraints at once.

01

Not a greenfield build

the EMR already had a live backend, data models, and workflows. With everything handled in-house, design and engineering were tightly coupled."

02

Doctors work in time-sensitive, high-stress settings and resist tools that disrupt their workflow

even small friction causes rejection."

03

The EMR's structure wasn't the problem

manual form-filling was

THE approach

Given the technical and behavorial constraints, the approach focused on reducing ambiguity early, validating feasibility fast, and grounding decisions in market reality and working prototypes.

  1. Defining Where AI Actually Helps

Before designing anything, we defined where generative AI could add real value reducing form-filling without touching the EMR's structure.

Before designing anything, we defined where generative AI could add real value reducing form-filling without touching the EMR's structure.

What We Did

  1. Started with the core hypothesis: AI can reduce documentation effort

  1. Started with the core hypothesis: AI can reduce documentation effort

  1. Mapped existing EMR fields and workflows to find repetition and context switching.

  1. Mapped existing EMR fields and workflows to find repetition and context switching.

  1. Translated these moments into clear, bounded AI use cases

  1. Translated these moments into clear, bounded AI use cases

Use cases

Auto-prefilling structured fields from typed or spoken input

Auto-prefilling structured fields from typed or spoken input

Detecting when a new symptom or diagnosis required a new EMR component

Detecting when a new symptom or diagnosis required a new EMR component

Contextual understanding basis doctors past and background for better templates

Contextual understanding basis doctors past and background for better templates

  1. Competitive Benchmarking & Market Understanding

competitor analysis

To ensure the solution addressed real gaps, not just internal assumptions.

To ensure the solution addressed real gaps, not just internal assumptions.

Competitive Benchmarking

Clear differentiation:

Most tools generate notes. This project focuses on generating the EMR itself.

  1. Prototyping Through Working Proofs, Not Static Screens

Real-time AI behavior couldn't be shown through static mockups - so we shifted to functional proofs of concept instead of screens.

tools used

Google AI Studio for rapid AI experimentation and POCs

Claude + Cursor to build and test a functioning prototype that reflected real EMR behavior

Let's see the timeline of this product coming to life

Let's see the timeline of this product coming to life

Timeline

Whiteboarding

The best way to get our thoughts down

Our initial approach was an entirely new platform integrated with our database.

Our initial approach was an entirely new platform integrated with our database.

  • An entry point, chat-first interface which acts like your personal asistant

    An entry point, chat-first interface which acts like your personal asistant

  • Navigational tabs, keeping a familiar interface to the original left panel navigation

    Navigational tabs, keeping a familiar interface to the original left panel navigation

  • Journey to the chat based EMR

    Journey to the chat based EMR

patient overview

detailed record about the patient

!! But this conflicted with our own principle, improve, don't rebuild, and raised real feasibility questions.

!! But this conflicted with our own principle, improve, don't rebuild, and raised real feasibility questions.

The Generative EMR as a feature

So we refocused: enhance the existing EMR and let doctors capture information as fast as possible.

So we refocused: enhance the existing EMR and let doctors capture information as fast as possible.

Voice and chat input where doctors can record sessions and the EMR automatically fills up based on the transcribed text.

Smart fill with components that are not mentioned being added to the EMR

Once the foundational generative EMR prototype was in place and internally tested, the focus shifted toward personalization and making the EMR smarter and more context-aware.

Once the foundational generative EMR prototype was in place and internally tested, the focus shifted toward personalization and making the EMR smarter and more context-aware.

Personalisation and context awareness

Next: making it personal.

  • Onboarding reads a clinician's background from the existing doctor database

  • Generates an editable starting template automatically

  • Doctors control which components appear, and how many EMRs they need

Next: making it personal.

  • Onboarding reads a clinician's background from the existing doctor database

  • Generates an editable starting template automatically

  • Doctors control which components appear, and how many EMRs they need

  • We used 4 doctors data to test out the personalisation

    We used 4 doctors data to test out the personalisation

  • First time user

    First time user

  • A starting EMR template basis the user, in this case a cardiologist

    A starting EMR template basis the user, in this case a cardiologist

  • A fully editable layout

    A fully editable layout

  • Edit using chat, voice or manually

    Edit using chat, voice or manually

the 3 ways a doctor will interact with the platform

Before and after

Varun's Portfolio

Final designs

Chat Bot

Texting anything can help fill fields, add new fields and finish the EMR faster

Just Speak!

Entire sessions can be recorded to fill fields in the fastest way possible

Key learnings

This was a the first project where I implemented an AI first feature

End-to-End Ownership

My first project run almost full-time taking on PM-adjacent scope, technical understanding, and close collaboration on implementation. It sharpened how I balance vision, feasibility, and execution.

My first project run almost full-time taking on PM-adjacent scope, technical understanding, and close collaboration on implementation. It sharpened how I balance vision, feasibility, and execution.

Static Designs Are Insufficient for Intelligent Systems

Designing AI-driven products means moving past screens and flows to understand behavior and real-time responses. Functional prototypes became essential, not optional.

Designing AI-driven products means moving past screens and flows to understand behavior and real-time responses. Functional prototypes became essential, not optional.

The Real Shift Is Structural, Not Cosmetic

In systems like EMRs, real innovation comes from restructuring how information flows not visual polish. Fixing the workflow mattered more than redesigning the interface.

In systems like EMRs, real innovation comes from restructuring how information flows not visual polish. Fixing the workflow mattered more than redesigning the interface.

Back to Home

Back to Home

Thanks for scrolling till here

Next Project

Next Project