Myntra | Aashay Ibrampurkar

2 MONTHS

CORE SYSTEM DESIGN, AI WORKFLOW DESIGN

INTERSHIP

Myntra

Designing an Indian vernacular language shopping experience & experimenting with an AI-ready Design System.

Throughout the internship I got to help all around the app. But majorly, I worked on two problems.

The first was winning and serving shoppers in Tier 2 and 3 cities. I built an interface translation framework for a new vernacular app experience, which grew out of my redesign of onboarding.

The second was a siloed & under-construction design system. I expanded it by coordinating with the entire design team, then structured it to scale for AI-enabled workflows.

Buy Now

अभी खरीदें

Native
Design System

CAUTION CAUTION CAUTION CAUTION CAUTION CAUTION CAUTION CAUTION CAUTION CAUTION

CAUTION NDA CAUTION NDA CAUTION NDA CAUTION NDA CAUTION NDA CAUTION NDA

Building the Vernacular app experience

(WHY VERNACULAR?)

Myntra’s next customers don’t shop in English

Myntra mostly grew up with shoppers in India’s big cities, who are comfortable with an app in English.

Myntra’s next wave of consumers live in smaller cities and towns, where almost everyone online uses an Indic language

M2 Users

Myntra’s new targeted group of users, coming primarily from Tier 2/3 cities.

98%

of India’s internet users have interacted with it in Indic Languages.

(A COLLOQUIAL ALTERNATIVE)

Plainly machine translating every word is not desirable.

English

Hindi

Sale ends tonight!

बिक्री आज रात समाप्त होती है

Too formal, not spoken in everyday speak,

Sale आज रात खत्म!

Closer to vernacular speak. Code mixed, informal, transliterated.

This leads to major issues for users who have limited comprehension in English.

Trust

a return policy the user cannot read.

Comprehension

a size guide or fabric description.

Friction

filter and navigation labels.

CORE DESIGN PROBLEM

How do you design a vernacular translation system?

We had to ask several questions while designing a system that needs to be smart, adaptable, self improving.

(SOLUTION SPACE)

How do you measure and define code mixing patterns?

Where is it most crucially needed in an app this big?

How does it get smarter once it’s live? Can it adapt to every individual user?

Unfortunately an NDA prevents me from showing you the solution in public domain, but i can always dive deeper when we chat…

CAUTION CAUTION CAUTION CAUTION CAUTION CAUTION CAUTION CAUTION CAUTION CAUTION

CAUTION NDA CAUTION NDA CAUTION NDA CAUTION NDA CAUTION NDA CAUTION NDA

Building an AI-ready Design System

(THE CONTEXT)

The product’s Design System was rapidly changing.

MDL 2.0

Fabric

While it was being rebuild, 10+ Product verticals (charters) had already built their own versions of the same components. Some screens had moved to the new system. Others hadn’t.

All of this while only the primitives and base components had yet been formally established.

Built inconsistently across charters

(THE MAIN JOB)

Auditing and defining these segregated components and organisms into Fabric.

The nuances of the task existed in maintaining consistency with the design system and merging only the components that were fit to scale throughout the global product.

Untangled

files from every team’s designers, auditing all new components

Decided and Merged

which components earn a place in the global system,

Extended

the foundations and organisms

Reviewed

developers’ builds against the design

Code Connected

with Figma’s component library

Other orgs like Atlassian, Microsoft, etc. have already connected their coded design system libraries using MCPs

But there were a couple hurdles to get over before Myntra could implement this solution

Under construction

The repository was still being finalised, so sandbox controls or even MCP for designers was a long way off.

Varies across charters

The system must become more granular and replicable to accommodate each charter’s local components.

(WHAT MAKES A TRULY EFFICIENT DESIGN SYSTEM FOR AI?)

CORE DESIGN PROBLEM

So, as an experiment, I tried building our own AI-ready design system.

I’m not a developer, and there was no sandbox to borrow, so I built a stripped-down system of my own and tested what it takes for an AI to choose the right piece, and build a delightful screen,

I built it bare on purpose and added one layer at a time, so I could see what each layer actually bought. Here’s a sneak peak of the questions i had to design for:

(SOLUTION SPACE)

Which layers should AI see or control, and when?

How does a team accommodate its own components?

What should AI do when nothing in the system fits?

How do you feed an AI detailed context and design instinct without burning tokens?

Unfortunately an NDA prevents me from showing you the findings in public domain, but i can always dive deeper when we chat…

(THE SYSTEM ISN’T ROBUST ENOUGH)

What if there isn’t a Figma frame to begin designing with?

Myntra was moving to AI-enabled design and development. Figma code connect serves one case well: a Figma frame exists and AI turns it into code, for further iteration.

I was curious about the other cases: quick iteration with no frame, a screen from a prompt, a new component from scratch.

(WHAT ELSE?)

Also got to work on a couple other shorter project sprints

Revamped onboarding

Aimed to convert shoppers with as little friction as possible

Conversational shopping

A hypothetical WhatsApp native shopping flow to target new audiences

Image asset generation

Assets for categories that are legible at small sizes

(IMPACT)

Small but (personally) valuable

Picked up.

The self-initiated vernacular project was taken on by PMs to build, launch and test in one part of the app.

20+ Components

refined and merged into the design system. Also audited, reviewed the code of many more.

(LEARNINGS)

Couldn’t have asked for a more diverse experience :’ )

Building for AI

Came in using AI and left able to build for it: rule files, skills and context an agent actually follows.

Small first

Start with the smallest version, then add one layer at a time, so you can see what each one changes.

Not just usability

Design for business outcomes and technical constraints together, by working with PMs and developers.

Let’s have a short chat?

to know more about me or the project…

Get in Touch