Omkar Khadamkar

Omkar Khadamkar

Staff UX Designer

AI-generated UI looks finished long before it is correct. I make design systems machine-readable — and I measure whether it worked. Most of the platform is built by people who don’t report to me.

Two renders of the same interface from a controlled field test. The left is labelled values left to inference, 93.2 percent; the right is labelled values named, 100 percent. Both look finished.
Same spec, two renders. Both pass a glance — one fails an audit. The field test →
Based Pune · remote or relocation Domain Enterprise security UX Patent First-named inventor Open to Principal IC · Design technologist · ManagerAvailable ImmediatelyCode github.com/omkarpkh
01Selected work
Shipped system
Running a design system an AI writes againstEnterprise security platform · 2022–2026

A component library got generated UI a quarter of the way onto our system. Documentation a machine couldn’t misread took it to 90%. Read the case study →

~2 hrs → 15 min correction−22% design time per screen25% → 90% on-systemMethod public, system confidential
 Shipped product
32 locations, one viewVMware NSX · 8 weeks · 2021–22

I argued the layout had to be replaced — then had to beat the concept my own team preferred. 43 participants picked mine. Now a filed US patent. Read the case study →

+30% completion−26% mis-clicks−155s workflowWorking recreation
 Field test
A rule without a value is not a rule01 · 2026

I published a claim about why AI-generated UI drifts, then built a rig to test it. The claim failed. I published that too. Read the case study →

93.2% → 100%Completeness +6.8 ptsStyle 0.0 ptsTwo failure modesSource public
 Field test
One example is not documentationFollows field test 01 · 2026

If naming every value is what matters, can one perfectly documented component teach the rest? It taught nothing — and made every remaining error deterministic. Read the case study →

Null result, publishedVariance → zero6 runs · 58 propertiesSource public
 Craft piece · Bulk edit
Twelve selected, twelve different valuesModel, interface and tests · 2026

Twelve rows selected, twelve different values — what should the field show? A working component you can use in the page, not a screenshot of one. 53 tests. Read the case study →

3 evaluations → converged6 directions53 tests, source publicLive wireframeBias disclosed
 Practice · Quality without authority
Nobody had to do what I said2013 – 2026

I have never had a direct report. Since 2013, getting design quality out of people under no obligation to give it to me — and what you have to build when you cannot simply decide. Read it →

0 direct reports, ever11 vendors, 4 cities5 designers on the systemGap stated first
Writing & research — where the findings started
AI drift — the field notes AI builds screens that look finished and aren’t. Five chapters on why it happens, how to write documentation a machine can’t misread, and which team catches which mistake. Twelve practitioners credited by name. Interactive · 5 chapters “AI drift, actual size” The original observation that started all of it: generated UI looks finished, and what is missing is never layout — it is precision. LinkedIn · 2026
02Practice

AI-native design systems

Component guidelines a model cannot misread, specs carrying exact values, and stylesheets bound to the production token file.

The system →

Drift verification

Regeneration diffs, audit rubrics, checksum tripwires. Proving a system still holds after AI touches it, rather than assuming it does.

The field test →

Systems at scale

Governance, adoption, contribution models, coaching. The unglamorous half: getting the thing used by people who did not build it.

How that works without a reporting line →

Design engineering

Component-to-code mappings, instrumentation specs, agent tooling. Built for engineering, for design, and once for a revenue team — only one of those is my own function.

A working component, 53 tests →
03Experience
YearsOrganisationRole & focus
2022 – 2026Skyhigh SecurityStaff UX Designer. Enterprise design system across SSE, ZTNA/Private Access, Cloud Firewall, CASB and reporting. AI-native documentation, drift verification, and agent tooling built for teams outside design.
2020 – 2022VMwareMember of Technical Staff (UX Designer). NSX networking and security. First-named inventor on VMware’s filed patent application for network status visualisation (US 2025/0036264).
2020FactSetSr. UX Specialist. Visual design guidelines across the product portfolio.
2019 – 2020ValuelabsSr. UX Designer. Enterprise workflows and component libraries. Design for Social Innovation award, UX Design India.
2013 – 2016Future GroupAssistant Design Manager. Guideline systems for manufacturers and buyers.
04Education
2023 – 2024UT Austin, McCombs School of BusinessPost Graduate Program, AI & Machine Learning
2019IIT RoorkeeDoctoral coursework — Inclusive Design. Completed the first semester, then left for a senior design role.
2016 – 2018National Institute of DesignMaster of Design
2009 – 2013National Institute of Fashion TechnologyBachelor of Design

Currently open to conversations

Design systems, AI-native documentation, and the seam where generated UI meets a real system. Remote roles and consulting both.

omkarpkh@gmail.com LinkedIn Questions you might have Field notes Résumé