Hello there, I'am
Amit Marathe
Principal UX Designer
Designing AI experiences that humans trust
17 years designing complex enterprise products across fintech, healthcare, and SaaS. Today I focus on what's next- AI agent workflows, human-AI collaboration models, and design systems that scale across product lines.
17+
Product Discoveries
+38%
Discovery > Engagement
14+
Enterprise Applications
4
Enterprise Design Systems
AI
As Productivity Multiplier &
AI Native Product Thinking
My recent articles
I design the space
between human intent
and AI capability.
Most AI products get the model right and the experience wrong. I focus on the layer that makes AI feel trustworthy, controllable, and genuinely useful — not just impressive.
The best AI experience
is the one where
human feels
More capable
I came to AI design through 17 years of enterprise UX — designing contract management platforms, compliance ecosystems, banking systems, and developer tools.
In every one of those domains, the biggest UX challenge was never the interface. It was cognitive load: too many decisions, too much information, too little clarity about what matters now. AI, done well, is the answer to that challenge. Done poorly, it's another layer of noise. My job is to make sure it's the former.
"The best AI experience is one where the human feels more capable - not replaced,
not overwhelmed, not dependent.
The AI should feel like a highly competent colleague who knows when to act, when to ask, and when to step back."
Currently, I am focusing on two capabilities
AI as a Productivity Multiplier
"How AI accelerates and improves traditional product design activities
AI-Native Product Thinking
"How I design AI agents,
AI experiences, and
human-AI collaboration models."
Design AI-Powered Compliance Platform
AI in practice: Enterprise Application
FDA compliance platform — platform modernisation with AI Registrar Corp × Coforge. Redesigned a fragmented compliance ecosystem unifying dashboard, marketplace, AI insights, and task management into one cohesive platform.
8
Distinct AI features
designed
38%
User registrations
in 2 quarters
+34%
Service subscriptions
30K+
Customers across
190 countries
Smart Onboarding with AI
AI search
Smart Compliance with AI
AI chatbot
Document Review with AI
Data Validation with AI
Image Editing with AI
Smart Insights with AI
Smart Onboarding with AI
Designed "Registro" — an embedded AI assistant that guides new users through FDA registration with progressive disclosure, smart defaults, and contextual explanations. Reduced onboarding time by 40% by replacing a multi-step form maze with a guided, conversational flow.
Conversational UX
Trust design
Document review with AI
Designed an AI-powered document validation layer that flags missing data, inconsistencies, and compliance risks before submission — replacing manual review with inline, contextual guidance. The key design challenge: making AI flags feel helpful, not alarming.
Error Prevention
Compliance UX
Smart compliance insights
Designed role-based AI insight cards that surface predictive compliance risks: expiring registrations, supplier risk scores (RegiScore), and market trend signals. The challenge was distinguishing "AI-generated" from "confirmed" data — I designed a clear confidence and source layer for every insight.
AI transparency
Data Vizard
AI search
Designed a natural language search experience across suppliers, products, regulations, and news — replacing a fragmented keyword search with intent-driven results. Designed the result ranking rationale UI so users could understand why results appeared, not just what they were.
Natural Language UX
Explainability
Image editing with AI
Designed AI-assisted document image correction flows — auto-alignment, quality checks, and redaction suggestions for compliance document uploads. Focused on undo-first design: every AI correction is a suggestion, never an automatic change.
Human override
Document UX
Data validation with AI
Designed inline AI validation for compliance forms — detecting invalid FDA codes, mismatched product categories, and missing regulatory fields in real time. The core UX challenge: validation that informs without interrupting the user's flow.
Error Prevention
Compliance UX
Smart compliance
monitoring
Designed an AI-driven compliance monitoring dashboard with deadline predictions, risk trajectory visualizations, and automated alert triggers (30/7/1 day reminders). Users stay compliant without needing to actively track — the system anticipates and surfaces what matters.
Proactive AI
Dashboard Design
AI chatbot — "Registro"
Designed the full conversational UX for Registro, the platform's embedded AI assistant — covering intent recognition flows, disambiguation patterns, graceful fallback states, and escalation to human support. Designed to feel like a knowledgeable compliance colleague, not a scripted FAQ bot.
Conversational Design
Agent UX
How I evaluate
every AI interaction
I design?
Before any AI feature goes to wireframes, I run it through this five-question framework. It's how I decide what the AI should do, how much autonomy it should have, and what the human always controls.
How I build a design system?
Building Design System for a large enterprise:
Aderant is a global leader in business management software for law firms. With the launch of a new cloud platform, they faced the challenge of unifying a diverse portfolio of products — each with its own UI patterns, built by distributed teams across multiple tech stacks — into a single, consistent, scalable experience. I led the design system from strategy to component rollout.
10+
Distributed designers
across globe
100+
Product design teams
3
UI stacks
150+
Components, tokens and templates
Faster time-to-market via reusable templates
Reduced design-dev training costs
Centralised updates
fewer regressions
Better cross-team collaboration
A design system is not a component library. It's a product that your designers and engineers use every day. I design it that way — with governance, adoption strategy, documentation, and a contribution model — not just a Figma file.
I don't just build components.
I build the systems that let organisations design at scale.
Every design system I've built has followed this sequence — adapted for the organisation's maturity, team size, and tech stack. The phases aren't always sequential; at Icertis and RegistrarCorp I ran discovery and pilot delivery in parallel to show early ROI while strategy was still being finalised.
Phase 1
Audit & inventory
UI audit of all existing screens. Component catalogue. Inconsistency map. Identify quick wins vs foundational needs.
Phase 5
Documentation
Usage guidelines, do/don't examples, prop specs, decision rationale. Supernova or Zeroheight. Engineers and designers as co-authors.
Phase2
Foundations
Tokens first: colour, typography, spacing, radius, elevation. Brand alignment. Dark/light mode strategy. Naming conventions.
Phase 6
Pilot adoption
Select 1–2 high-visibility teams for first rollout. Learn what breaks. Fix before scaling
Phase 3
Core components
Atoms → molecules → organisms. Figma library + Storybook in sync. Accessibility built into every component, not retrofitted.
Phase 7
Scale & evangelise
Design showcases, internal workshops, before/after demos. Adoption isn't passive — it's marketed internally.
Phase 4
Governance
Contribution model, approval workflow, versioning policy, deprecation process. Define before the first PR arrives.
Phase 8
Measure & iterate
Component adoption rates, version drift, time-to-integrate, UI inconsistency regression. Treat the DS like a live product.
How I think about design systems?
Most design system projects fail not because the components are wrong, but because no one designed the system around the people who have to use it. Engineers skip it because it's not in their stack. Designers bypass it because it's easier to build from scratch. PMs deprioritise it because the ROI is invisible until it isn't.
I've built four enterprise design systems across different organisations, domains, and team sizes. Every time, the hardest challenge wasn't the token architecture or the component API — it was adoption. Getting 100 engineers across 3 tech stacks to use the same button is a change management problem as much as a design one
1
Design systems are products
Treat them like one. They need a roadmap, a versioning strategy, a contribution model, and adoption metrics. A Figma file is not a design system.
4
Figma ↔ Code parity matters
The distance between a Figma component and its code equivalent is where inconsistency lives. I work closely with frontend leads to keep Figma and Storybook in sync — not as a handoff, but as a shared source of truth.
2
Governance before scale
The moment more than one team uses a design system, you need governance. Who can contribute? Who approves changes? What happens when a component breaks? Define this before you have a crisis
5
Documentation is a design deliverable
Every component needs usage guidelines, do/don't examples, accessibility notes, and a decision rationale. Documentation that lives only in designers' heads doesn't scale.
3
Tokens are the real foundation
Components are visible. Tokens are structural. A well-defined token architecture — spacing, colour, radius, elevation — is what makes a design system actually portable across products and themes.
6
Adoption is designed in
I design the onboarding experience for the design system itself — how do engineers discover components? How do designers contribute? Adoption is not a training problem. It's a UX problem.
Taking idea to shipped mobile app
Progressive mobile application for B2C
The fastest path to a credible MVP isn't more features, it's a tight loop of ecosystem research, a single validated core action, and instrumented testing from the first release.
Talento Today went from founder vision to a successfully validated beta by treating research and analytics as part of the design process, not a step before or after it.
30 M
User base
3000+
Users in the
first 6 months
#2
Sports App in
Germany
What it actually takes to go from a founder's idea to a validated, beta-released consumer app condensed from designing Talento Today as sole product designer for an early-stage German startup
5 days
to map the entire ecosystem before any design work
12
User interviews across 3 countries before wireframes
3
Distinct personas designed as one connected system
1
designer, full ownership: research to visual design
Phase 1
Map the ecosystem
Workshop with founders + eng lead. Roles, workflows, pain points \u2014 before any screens
Phase 2
Validate with users
12 interviews, real personas, 4 competitor products benchmarked
Phase 3
Design the core loop
One central action across all personas. Reusable components from day one
Phase 4
Test, measure, ship
On-field testing + analytics before and after beta release
