Medical AI - ambient - scribe
DocAssist: Rethinking clinical conversations with GenAI
Built a dual-assist AI that lets doctors query medical knowledge and record consultations—powering real-time suggestions and clinical notes.
Kick off Medical AI assistance in a tap
Easily initiate DocAssist directly from your queue with full patient context, or in a blank session when needed.
Talk, type, or both: Get help in the moment
Chat with the AI while recording your consultation. Get real-time prompts, insights, and follow-up questions as the visit unfolds.
From conversations to clinically useful notes
Unlock contextually aware notes with dynamically integrated data from previous patient encounters, health system-specific guidelines, and clinician preferences.
👥 My Team
3 product designers
2 development teams
👨‍💻My Role
Design clinic consultation workflows driven by DocAssist for Eka Doc Mobile App, and lead EkaScribe chrome extension experience.  
⏳ Duration
Jan 2025
Background
While our DocAssist chat was already live on Eka Doc's web tool, we wanted to dig deeper into real-world feedback. We connected with our top users—their inputs were clear: doctors needed faster, smarter ways to document consultations or structured clinical notes in general. A recurring theme from earlier research resurfaced: the ability to generate prescriptions directly from conversations.
“Aap consultation sun kar parcha nai bana sakte?”
“Can’t you generate prescription by listening to consultation?”
With our company's strategic shift toward GenAI, our Data Science team developed an AI scribe capable of transcribing consultations and creating structured clinical notes (and prescription). A small MVP showed strong adoption, giving us confidence to scale the feature, along with DocAssist Chat, to all mobile users.
Discover
Discovery
Understanding the gap
While analyzing AI chat usage patterns on our web tool, we found strong adoption—doctors actively used it as a medical assistant and "Google for consultations". Alongside, insights from earlier shadowing studies of real-world consultations surfaced challenges that still held true:
  • Doctors started visits by gathering patient vitals, medical history, and past records - all from various sources.
  • Mid-consultation, they constantly toggled between Eka Doc tool and patient, breaking eye contact, flow, and patient trust.
  • Different labs reporting varying normal ranges made it harder to quickly interpret results, forcing doctors to cross-verify online.
  • To explain concepts, they often showed Google results to patients directly.
Every switch between patient and screen created tiny moments of friction—moments where attention drifted, connections weakened, and care became mechanical.
The India factor
Designing for Indian healthcare meant adapting to unique ground realities:
  • Regulatory pressure around detailed prescriptions is low; in busy OPD (Outpatient Department) settings, doctors often have only a few minutes per patient—leading to a focus on rapid diagnosis over comprehensive documentation.
  • To save time, doctors frequently adopt personalized shorthand and informal writing styles, with minimal regulatory oversight or standardization.
A conversation with Dr. Deepika Ponnappa, Co-founder of Healthland Clinics, captured it best:
“Nowadays it’s hard to find good quality Rx (prescription) in India. Doctors prescribe medications and lab tests but often skip the details. I want every Rx from Healthland to be something any person can read and clearly understand the full picture!”
Defining AI design principles
I wanted to define key principles that will guide all the design and product decisions for DocAssist - to ensure we always align with our vision and provide a consistent user experience.
Contextual
Understand patient’s context and doctor’s intent, and assist accordingly - deliver relevant, real-time prompts - adapted for Indian medical practices.
Flexibility
Doctors must be able to move effortlessly between talking, querying AI, and clinic workflows without friction.
Speed & Clarity
Deliver information quickly without slowing consultations, always presenting a clear and usable output.
Grounded Reliability
AI responses should stay closely tied to real-world medical facts, doctor inputs, and the ongoing conversation. No hallucinations, assumptions, or unrelated information
Invisible Assistance
AI should remain in the background—enhancing, not interrupting, doctor-patient conversations.
Transparency
Clearly show AI’s actions — what it heard, how it interpreted, and how it generated notes. So doctors remain in control and can easily verify or edit information.
Designs
DocAssist on mobile
Building on the insights from discovery and guided by our design principles, I crafted a new experience for DocAssist AI within the Eka Doc mobile app.
‍
Early explorations included recording conversations and chat as a separate features, but this felt limiting. Revisiting the guiding principles, I realized that conversation recording and chat should work together — enabling doctors to query medical information in real time, receive contextual suggestions, and reduce their dependence on external searches like Google.
Eka Scribe chrome extension
To make documentation even faster, we built Eka Scribe — a Chrome extension that listens to consultations and auto-generates structured clinical notes in formats such as EMR pad and SOAP.
‍‍
We also designed clear setup flows to guide doctors through the extension. Since the LLM model is still evolving, every session emphasized input/output selection (such as preferred languages and output templates) — ensuring quick setup and real-time adaptability based on the patient.
Impact and Next steps
We successfully launched DocAssist and Eka Scribe, driving rapid adoption — with 15% of Eka's daily active users (DAU) engaging with the new AI tools within just one month of launch.

Building on this momentum, the business team plans to expand DocAssist to the US market in the coming months.
Audit based on Guidelines for Human-AI Interaction
After a rapid sprint to bring the first version of DocAssist AI to life, we took a step back to audit the experience against our AI design principles and Microsoft's HAX Toolkit.
This audit was not a retrospective, but a forward-looking exercise to identify key next steps for improving reliability, trust, and usability.
Next steps
Grounding and tool calling
In next few weeks, keeping our AI principle of Tranparency in mind,  we plan to ship front-end experience around making the grounding and tool calling capabilities explicit for the doctors. With this release, doctors will be informed of what tools have been invoked to ensure minimum hallucinations, assumptions, or unrelated information in the responses. They will also be able to view the sources if needed.
Onboarding for DocAssist
Instead of throwing doctors to the chat with little context of what to do, we can help them understand how AI can help. With the limited learning we have, an onboarding carousel can be a good start.
Other work
Accomodation - Mobile app
Hosting on MakeMyTrip
Crafting hosting experience for India's leading online travel company and help hosts reach out to 50M+ guests on MakeMyTrip & its partner websites.
Read case study

Let’s grab a ☕ and chat!

Always open to interesting conversations, collabs, or anything in between - let’s build something meaningful together.
‍
Feel free to email me or reach out on Linkedin.
Made with ✨ by Suyash