Umesh D. Anghan — AI-Driven Product Designer

Portfolio

Selected AI-powered projects

In-depth case studies, Figma designs, live AI-powered products, and visual portfolio showcasing work across healthcare, fintech, legal tech, enterprise and more — delivered using AI-augmented design workflows.

nTouch

SSE — Sports Stock Exchange

Royal Oman Police (ROP)

nivanophysicians.com ↗

Healthcare

Healthcare

Healthcare

apollomanagedcare.com ↗

Healthcare

Visual portfolio thumbnails — website, mobile, web app, and other design work.

WebsiteQICDRC

WebsiteAAH

WebsiteMMQ

WebsiteSurge

Websitebeáh

WebsiteVcare

WebsiteSocial Connect

WebsiteCodemaster

WebsiteVastu

WebsiteDulles

WebsiteMG Hosting

WebsiteAllergy EHR

WebsiteDr. Paul

WebsiteNivano Physicians

WebsiteNetwok

WebsiteNJ Experts

WebsiteCosmetisuite

WebsiteNorthVally Family Physicians

WebsiteDr Catalyst

WebsiteWebsite4MD

WebsiteImmunology Specialist

Mobile AppSSE — Sports Stock Exchange

Mobile AppMyBZZ

Mobile AppOffice Mato

Mobile AppNetwok Mobile

Mobile AppCosec Apta

Mobile AppPayQuick

Mobile AppEngage

Mobile AppJob Portal

Mobile AppStudy Science

Mobile AppQuick Realtor

Mobile AppPizza Stand

Web AppInsurance Portal

Web AppStock Dashboard

Web AppOffice Mato Web

Web AppDefine X

Web AppEngage Web

Web AppCustom Dashboard

Web AppCosec

Web AppSAMAS

Web AppDR Catalyst

Web AppShowItBig

Web AppBooking System

Visual DesignVisual Design — Brand Identity

Visual DesignVisual Design — UI Components

Visual DesignVisual Design — Icon Set

Visual DesignVisual Design — Illustration

Visual DesignVisual Design — Marketing

Enterprise SaaS · Healthcare

Claim Management Software

End-to-end UX redesign of a complex insurance claim processing platform for healthcare providers — reducing claim rejection rates and accelerating reimbursement cycles.

RoleLead Product Designer + Scrum Master

Duration6 Months

PlatformWeb App (Desktop-first)

ToolsFigma · FigJam · Claude.ai · Storybook

↓22%Claim rejection rate

57%Faster submission time

81CSAT score (was 54)

96%Task completion rate

The Problem: Billing specialists faced a 34% claim rejection rate driven by coding errors, a fragmented 11-click submission workflow across 3 disconnected legacy screens, and zero real-time validation feedback — costing healthcare providers millions in delayed reimbursements annually.

01 · Research Understanding Claims Chaos

12 stakeholder interviews, VOC analysis, Claude.ai transcript synthesis

40+ hrs interviewsVOC analysis

02 · Define Framing the Problem

3 personas, journey mapping, HMW statements, impact–effort matrix

3 PersonasIA mapping

03 · Design Unified Claim Workspace

Single workspace, real-time validation, 80+ component design system

Figma MakeStorybook

04 · Prototype 5 Core Flows

Figma hi-fi prototype + Lovable live demo for exec stakeholders

Lovable3 iterations

05 · Testing Validated Outcomes

8 billing specialists, 2 moderated sessions, post-launch metrics

Usability testingCSAT tracking

01 · Research

Understanding the Claims Chaos

Conducted 12 stakeholder interviews with billing specialists, practice managers, and insurance coordinators. Used Claude.ai to synthesise 40+ hours of interview transcripts into structured insight clusters. Ran VOC analysis to identify top pain points: a 34% claim rejection rate due to coding errors, a fragmented multi-screen workflow requiring 11 clicks per submission, and zero real-time validation feedback.

34%Avg. claim rejection

11Clicks per submission

3Disconnected screens

0Real-time validation

02 · Define

Framing the Problem Space

Created 3 user personas — the Billing Specialist, the Practice Manager, and the Compliance Officer — each with distinct goals and frustrations. Mapped a current-state journey exposing 6 critical failure points where claims fell through. Defined the design challenge: "How might we give billing teams real-time confidence at every step of a claim's lifecycle — from entry to adjudication — without requiring retraining?" Prioritised features on an impact-effort matrix with product and engineering leads.

3 User PersonasJourney MappingHMW StatementsImpact–Effort MatrixInformation Architecture

03 · Design

Redesigning the Claim Lifecycle

Consolidated 3 legacy screens into a single unified claim workspace with a contextual sidebar. Designed an inline ICD/CPT code validator with real-time error flagging before submission. Built a claim timeline component showing adjudication status at a glance. Used Claude.design and Figma Make to rapidly generate layout alternatives and validate information hierarchy. Delivered a full design system in Figma with 80+ reusable components documented in Storybook for seamless developer handoff.

Unified WorkspaceReal-time ValidationDesign System · 80+ ComponentsStorybook DocsWCAG 2.1 AA

04 · Prototype

Interactive Prototyping & Stakeholder Validation

Built a high-fidelity clickthrough prototype in Figma covering 5 core flows: new claim entry, claim edit & resubmission, batch processing, ERA reconciliation, and dashboard reporting. Used Lovable to create a live coded prototype for executive demos — enabling non-technical stakeholders to interact with the real UI in a browser. Ran 3 rounds of internal review with product, engineering, and compliance, iterating on the status workflow and notification system each cycle.

5 Core FlowsFigma PrototypeLovable Live Demo3 Iteration Rounds

05 · Testing & Outcomes

Usability Testing & Measurable Impact

Conducted moderated usability testing with 8 billing specialists across 2 sessions. Measured task completion rate, time-on-task, and error rate for the claim submission flow. Post-launch metrics showed dramatic improvement — claim submission time dropped from 4.2 minutes to 1.8 minutes, the rejection rate fell by 22 percentage points in the first quarter, and CSAT rose from 54 to 81. The redesign became the foundation for a platform-wide design system rollout.

↓22%Claim rejection rate

57%Faster submission

81CSAT (was 54)

96%Task completion

E-commerce · Life Sciences

E-commerce Platform for Life Science Products

Designing a dual-tier B2B purchasing experience for high-value laboratory instruments and low-value consumables — serving researchers, procurement managers, and lab technicians.

RoleLead UX / Product Designer

Duration8 Months

PlatformWeb Responsive · B2B E-commerce

ToolsFigma · Claude.ai · Figma Make · Miro

+31%Quote-to-order conversion

+44%Repeat consumable orders

61%Faster approval cycle

4.6/5Usability score

The Problem: A life science distributor needed a single B2B e-commerce platform that could serve two radically different buyer types — researchers making 6-week, multi-stakeholder $500K instrument decisions, and lab technicians doing weekly $50 consumable replenishments. One UX couldn't serve both modes without breaking one.

01 · Research Two Buyer Modes

15 contextual inquiries, purchase behaviour analysis via Claude.ai

15 interviewsBehaviour patterns

02 · Define Dual Journey Architecture

4 personas, considered purchase flow vs quick reorder flow

4 Personas9 Design Principles

03 · Design Adaptive PDP System

Adaptive product pages, multi-approver checkout, 120+ components

Figma MakeRole-based UI

04 · Prototype 2 Parallel Prototypes

Separate journeys tested simultaneously with real user pairs

Scenario-basedQuote→PO flow

05 · Testing Dual Buyer Validation

11 participants across 2 buyer types, post-launch growth metrics

11 participantsPost-launch tracked

01 · Research

Two Buyer Modes, One Platform

Conducted contextual inquiry sessions with 15 users spanning research scientists, lab procurement officers, and department heads at life science organisations. Leveraged Claude.ai to synthesise purchase behaviour patterns and segment insights by product tier. Key finding: high-value purchases (microscopes, sequencers, centrifuges — $5K–$500K) involve a 3-to-6 person approval chain and 2–8 week decision cycles, while low-value consumables are replenished weekly by a single technician. These two modes required fundamentally different UX journeys within the same platform.

2Buyer modes

6 wksHigh-value cycle

15User interviews

3–6Stakeholders per purchase

02 · Define

Separating the Two Purchase Journeys

Defined four personas: the Researcher (explores, compares specs), the Lab Tech (quick reorder, volume buying), the Procurement Officer (approval workflows, budget tracking), and the Department Head (analytics, vendor management). Mapped two separate purchase journeys — a Considered Purchase Flow for high-value items with quote requests, comparison tools, and multi-approver sign-off, and a Quick Reorder Flow for consumables with saved lists, bulk add-to-cart, and subscription ordering. Documented 9 design principles guiding the dual-mode architecture.

4 PersonasDual Journey ArchitectureQuote Request FlowQuick Reorder Flow9 Design Principles

03 · Design

A Platform That Knows What You're Buying

Designed an adaptive product detail page that intelligently switches between a High-Value Mode (deep spec tables, document downloads, comparison drawer, quote CTA, demo scheduling) and a Consumable Mode (quantity selector, pack-size variants, reorder history, subscribe-and-save). Built a multi-approver checkout with role-based permission states — requesters, approvers, and budget holders each see a tailored view. Used Figma Make to generate and evaluate 12 layout variants for the product catalogue grid. Delivered 120+ Figma components covering both product tiers, account management, and the approval dashboard.

Adaptive PDPMulti-approver CheckoutSubscribe & Save120+ ComponentsRole-based UI

04 · Prototype

Simulating Both Purchase Journeys Simultaneously

Built two parallel Figma prototypes — one for the high-value considered purchase journey (browse → compare → request quote → multi-approver review → PO generation) and one for the Quick Reorder journey (saved list → bulk add → express checkout → subscription confirm). Ran scenario-based walkthroughs with a procurement manager and a lab technician simultaneously to expose conflicts in shared components (navigation, cart, account) that needed to serve both modes without confusion. Three rounds of cross-functional iteration followed.

2 Parallel PrototypesScenario WalkthroughsQuote → PO FlowExpress Checkout

05 · Testing & Outcomes

Validating Across Both Buyer Types

Ran two separate rounds of moderated usability testing — one with 6 procurement/researcher profiles for the high-value flow, one with 5 lab technicians for the reorder flow. Key improvements from testing: the comparison drawer needed a sticky anchor, the approval notification email needed an in-context preview, and the quick-reorder list needed stronger visual differentiation from the standard cart. Post-launch: high-value quote conversion increased by 31%, repeat consumable orders grew by 44%, and average procurement approval time dropped from 9 days to 3.5 days.

+31%Quote conversion

+44%Repeat orders

61%Faster approvals

4.6/5Usability score

AI Product · E-commerce · Life Sciences

AI-Based Agent for E-commerce Content Generation

Designing the UX for an AI-powered content generation agent that automatically produces product titles, descriptions, spec tables, and SEO metadata for 50,000+ life science SKUs — cutting content production time by 80%.

RoleLead Product Designer · AI Workflow Designer

Duration5 Months

PlatformInternal Web Tool · AI SaaS

ToolsClaude.ai · Figma · Figma Make · Lovable

16×Content production speed

71%AI acceptance rate

12KSKUs cleared in month 1

320SKUs per editor per day

The Problem: The life science e-commerce platform had 50,000+ SKUs with incomplete or missing product descriptions. At 20 descriptions per editor per day, full catalogue coverage would take 7 years. The challenge wasn't just automation — it was building enough trust in AI output that editors would use it rather than ignore it.

01 · Research The Content Bottleneck

Stakeholder interviews + Claude.ai synthesis, 3 key trust constraints identified

50K SKU gapTrust framing

02 · Define Human–AI Collaboration

3 personas, 4 workflow modes, transparency + control + auditability framework

Confidence scoringApproval gates

03 · Design AI Content Workspace

Split-panel workspace, confidence indicators, Tone & Template Studio, Batch Review Grid

Claude.design14 content fields

04 · Prototype Live AI Prototype

Lovable prototype with real Claude.ai API — editors tested actual AI output, not mockups

Live API3 sprint cycles

05 · Testing Trust Score Metric

10 editors, Trust Score primary KPI, 71% acceptance rate achieved

16× improvement12K SKUs cleared

01 · Research

The Content Bottleneck Problem

The platform had a critical content gap — over 50,000 product SKUs with incomplete, inconsistent, or missing product descriptions, causing low discoverability and high bounce rates. Through stakeholder interviews with content managers, catalogue editors, and product owners — supported by Claude.ai-powered transcript analysis — we found a single editor could produce ~20 quality descriptions per day. Full catalogue coverage would take 7 years. Three critical constraints emerged: scientific accuracy was non-negotiable, brand voice consistency was critical, and editors needed to trust and control AI-generated content rather than simply accept it.

50K+SKUs needing content

7 yrsManual completion time

20Descriptions/editor/day

3Critical trust constraints

02 · Define

Designing for Human–AI Collaboration

Defined three primary personas: the Content Editor (reviews and approves AI output), the Catalogue Manager (monitors batch progress and quality scores), and the Product Owner (sets brand voice, tone rules, and output templates). The core design challenge: "How might we make AI-generated content feel like a trusted first draft rather than an unreliable black box?" We defined a principle-led framework — Transparency (show confidence scores), Control (editable inline), and Auditability (track every human edit vs AI output). Mapped 4 workflow modes: single-item generation, batch generation, bulk review & edit, and publish workflow with approval gates.

3 PersonasHuman-AI Collab ModelConfidence Scoring4 Workflow ModesApproval Gates

03 · Design

The AI Content Workspace

Designed a split-panel AI workspace: left panel shows raw product data (SKU, attributes, existing specs), right panel shows the live AI-generated content with field-level confidence indicators (green/amber/red). Each generated field is independently editable with change tracking. Built a Tone & Template Studio where brand managers configure voice guidelines and output structure — feeding directly into the AI prompt pipeline. Designed a Batch Review Grid for processing 100–500 SKUs simultaneously with an accept/edit/reject triage system. Used Claude.design for rapid UI exploration of confidence scoring visualisations and Figma Make for component variant generation across 14 content field types.

Split-panel WorkspaceConfidence IndicatorsTone StudioBatch Review GridChange Tracking

04 · Prototype

Making AI Feel Tangible

Built a high-fidelity Figma prototype simulating the AI generation experience — including animated typing states, confidence score population, and inline edit interactions. Used Lovable to build a functional coded prototype that integrated with a real Claude.ai API endpoint, allowing content editors to interact with actual AI-generated product descriptions during testing sessions — not just static mockups. This was critical for measuring genuine trust and usability rather than reactions to a simulation. Ran the prototype through 3 sprint review cycles as Scrum Master, incorporating engineering feasibility feedback into each design iteration.

Animated AI StatesLive Claude.ai IntegrationFunctional Lovable Prototype3 Sprint Cycles

05 · Testing & Outcomes

Building Trust in AI-Generated Content

Ran 2 rounds of usability testing with 10 content editors using the live Lovable prototype. Primary metric was Trust Score — measured by how often editors accepted AI output without editing (target: >60% acceptance). Initial round revealed that editors distrusted amber-scored fields even when content was accurate — leading to a redesign of the confidence explanation tooltip with specific reasons for the score. Final round showed 71% average acceptance rate and a measured throughput of 320 SKUs per editor per day — a 16× improvement. Post-launch, the platform cleared 12,000 SKUs in the first month alone.

16×Production speed

71%AI acceptance rate

12KSKUs cleared month 1

320SKUs/editor/day

Date of Birth	19 December 1991
Gender	Male
Marital Status	Married
Location	Surat, Gujarat – 394105
Email	anghanumeshd8@gmail.com
Phone	+91 94272 01756
Portfolio	uxid.vercel.app ↗
LinkedIn	View Profile ↗
Dribbble	@Udanghan ↗
Behance	@udanghan ↗
Medium	@anghanumeshd8 ↗
Direction	Get Direction ↗

Designing
with AI,
for Humans.

Designing AI-powered products people love to use

Companies I've Worked With

My AI-first design toolkit

My tools

Where I've made an impact with AI & design

Educational Qualification

Selected AI-powered projects

Claim Management Software

Understanding the Claims Chaos

Framing the Problem Space

Redesigning the Claim Lifecycle

Interactive Prototyping & Stakeholder Validation

Usability Testing & Measurable Impact

E-commerce Platform for Life Science Products

Two Buyer Modes, One Platform

Separating the Two Purchase Journeys

A Platform That Knows What You're Buying

Simulating Both Purchase Journeys Simultaneously

Validating Across Both Buyer Types

AI-Based Agent for E-commerce Content Generation

The Content Bottleneck Problem

Designing for Human–AI Collaboration

The AI Content Workspace

Making AI Feel Tangible

Building Trust in AI-Generated Content

Have an AI Product to Design?

Designingwith AI,for Humans.

Designing AI-powered products people love to use

Companies I've Worked With

My AI-first design toolkit

My tools

Where I've made an impact with AI & design

Educational Qualification

Selected AI-powered projects

Claim Management Software

Understanding the Claims Chaos

Framing the Problem Space

Redesigning the Claim Lifecycle

Interactive Prototyping & Stakeholder Validation

Usability Testing & Measurable Impact

E-commerce Platform for Life Science Products

Two Buyer Modes, One Platform

Separating the Two Purchase Journeys

A Platform That Knows What You're Buying

Simulating Both Purchase Journeys Simultaneously

Validating Across Both Buyer Types

AI-Based Agent for E-commerce Content Generation

The Content Bottleneck Problem

Designing for Human–AI Collaboration

The AI Content Workspace

Making AI Feel Tangible

Building Trust in AI-Generated Content

Have an AI Product to Design?

Designing
with AI,
for Humans.