Lead / Staff Product Designer · 11+ years

I design the software that high-stakes work runs on.

Shipped for
  • Fannie Mae
  • PayPal
  • Vanguard
  • Atlassian
  • InMobi

I turn complex, policy-heavy workflows into clear products across enterprise, fintech, developer platforms and AI, where decisions, accountability and human override stay explicit. I prototype and build design-system components in React, TypeScript and Tailwind with Claude Code, Cursor and Codex, reviewed and shipped by engineering.

Read the MUSE case study →

Available now · Austin, TX · Remote, hybrid or onsite

Vinoth Kumar Manickam, Lead / Staff Product Designer, at work
  • 5 days → 2 daysmodel approval time (MUSE)Fannie Mae · MUSE
  • 62%Platform-reported, from about 200 to 300 errors a week. Post-launch count and period not published.fewer platform-reported errorsPayPal · developer portal
  • ~2 hoursApproximateto approve a claim, from about 2 weeks (up to 2 months at the slowest); error cases 2–4 daysFannie Mae · Claims & Recovery

A disagreement I lost, then partly won

Tension
Model-risk governance wanted all five validation gates to stay blocking, so no model could advance without a completed sign-off at every stage.
My call
I recommended that gates 1 and 2 become non-blocking with mandatory evidence capture, because validators and model-risk staff told us those gates almost never changed an outcome, they only delayed it.
Result
Gates 1 and 2 stopped blocking. For roles that must sign, another approver can now send a reminder, and some roles have a delegate who can approve on their behalf. Documents for approval were surfaced into the approval screen. Then an AI review rated each submission low, medium or high. Approval went from about 5 days to 2 days, and with the AI step to about 5 hours.
G1G2G3G4G5NON-BLOCKINGUNCHANGEDSEQUENTIAL LIFECYCLE
Read the full trade-off
Evidence I used
Twelve validator interviews, input from MRM staff, and a line-by-line read of the internal validation guideline to confirm nothing required those two gates to block.
The decision that was made
Gates 1 and 2 became non-blocking. The mandatory multi-role attestations in review approvals were kept and cannot be skipped.

Selected work

Four projects. Each states the official role, the team relationship, and how the outcome was measured.

Approximate
= rounded or estimated figure
Reconstructed MUSE model registry screen showing validation status, version history and open findingsFannie Mae
Shipped

Jun 2023 — Aug 2026

MUSE — enterprise model risk management

Model registration, validation and sign-off lived across disconnected tools, so no one could prove a model was reviewed, current and approved for its stated use.

Role · Lead Product Designer · sole product designer for MUSE, serving ~500 internal users who review and approve models

MUSE model approval time: 5 days → 2 days, about 5 hours with the AI review step; approval now runs in three stages

How this was measured
Measured from when Clearing House was merged into MUSE. Before: about 5 days to get a model through approval. After gates 1 and 2 stopped blocking, reminders and delegates were added, and approval documents were surfaced into the approval screen: 2 days. With the hybrid AI-assisted flow (an AI rating of each submission): about 5 hours. My contribution: the approval flow and interaction design; product and engineering built and measured it.
Reconstructed PayPal developer portal screen showing an integration path, credentials and webhook eventsPayPal
Shipped

Oct 2021 — Jun 2023

Developer portal and documentation experience

Developers stalled mid-integration because the portal, documentation and tooling each described the same task differently.

Role · Senior Product Designer · design lead for the portal and documentation experience

Approximate

Monthly U.S. developer-portal visits: 45K pre-launch baseline → 160K in the quarter after launch

How this was measured
Unit: monthly developer-portal visits — not registered developers and not monthly active developers. Geography: U.S. Baseline: 45K monthly visits pre-launch. Result: 160K monthly visits during the quarter after launch. Source: platform-team analytics. My contribution: design lead for the portal and documentation experience.
Authentic Jira Data Center administration artifacts: the cluster update-monitoring table during a mixed-state upgrade, the zero-downtime upgrade system flow behind it, and the Jira downtime exception dialogAtlassian
Shipped

2017 — 2018

Jira Data Center zero-downtime upgrades

Administrators dreaded cluster upgrades. A wrong compatibility call or an invisible exception could turn a 'zero downtime' upgrade into an outage, with no guided way to see readiness, monitor progress, or recover.

Role · UX Designer · primary designer for UX, UI and research

A guided administrator experience shipped in Jira Data Center spanning readiness, compatibility, live upgrade monitoring, exception handling and completion.

Sanitized line drawing of a button and data-table row with spacing, radius and token annotationsFannie Mae
Shipped

Jun 2023 — Aug 2026

AI-assisted design systems

Design decisions drifted between Figma, tokens and component code. I tested whether AI coding agents could keep production code faithful to design intent without weakening governance.

Role · Lead Product Designer · owned the design system

Explicit design-system rules, governed components and a human-reviewed path from Figma to Build, Test, Deploy.

More outcomes

  • Fannie Mae

    About 2 weeks → about 2 hours

    Approximate

    Human-in-the-loop AI claims inspection

    AI reviews contractor-submitted photos of roofs, repairs and structural damage and auto-approves qualifying submissions, with image-quality checks, retake flows, escalation paths, live-video fallback and human override for exceptions. Up to 2 months at the slowest before. Claims that return an error are inspected in 2–4 days; claims that pass the AI checks are approved automatically.

    Read the Human-in-the-loop AI claims inspection case study
  • Fannie Mae

    Recovery and billing: case and invoice processing went from about one week to a few hours.

    Approximate

    AI document extraction in recovery and billing

    This is my own before and after observation, not an instrumented measure.

Building in code

  • Stack

    React · TypeScript · Tailwind CSS · HTML · CSS · JavaScript

    Claude Code · Cursor · Codex · Figma Make · Lovable

  • Tested against live data

    I build working prototypes of approval stages, risk categories, form controls and data tables, and test them against live APIs. One data table crashed on real responses, an issue static mockups would not have shown, and it was caught before any engineering time was spent.

  • Design systems in code

    I built and updated design-system components in the codebase with AI coding tools under engineering review, and used AI agents to audit implementations against system standards, including WCAG 2.2 AA issues. I wrote agent instruction files so coding agents default to correct system usage.

    Read the case study →

How I operate

  • Led

    Design as the sole product designer

    Served as the sole product designer for MUSE, setting product design direction across the model lifecycle while partnering with product, engineering, validators and model-risk governance.

  • Recommended

    Strategy that changed a roadmap

    Recommended that PayPal treat integration delay as an information-architecture problem; product accepted it and redirected the portal roadmap.

  • Decided

    Product architecture and interaction direction

    Decided the MUSE information architecture and stage model. Model-policy approval stayed with risk governance — I did not own that decision.

  • Facilitated

    Alignment across risk, product and engineering

    Facilitated gate-ownership negotiation with model-risk governance and presented direction to Director, Senior Director and VP stakeholders.

What I did not own

Model-policy approval. That stayed with risk governance. I decided design direction and the functional definition of the platform. I presented direction to Director, Senior Director and VP stakeholders.

Four stages I run on every regulated or AI-enabled product

  1. 01

    Frame the system

    Map the policy, the roles and what the model is actually accountable for.

  2. 02

    Define accountability

    Name who decides, who reviews and who can override — before any screen exists.

  3. 03

    Design the human-AI decision

    Confidence, lineage and override paths staged for the moment of judgment.

  4. 04

    Measure the outcome

    Agree how success will be measured with product and analytics before launch.

About, briefly

I design for people who have to answer for their decisions. Most recently I was Lead Product Designer for MUSE at Fannie Mae, serving as the sole product designer on the platform.

  • Fannie Mae · Lead Product Designer · Jun 2023 — Aug 2026
  • PayPal · Senior Product Designer · Oct 2021 — Jun 2023
  • Vanguard · Senior UX Designer · Oct 2019 — Sep 2021

Hiring for a Lead or Staff product design role?

Based in Austin and available now. Happy to walk a hiring team through any figure on this site.

Austin, TX · Central Time · Available now · Onsite, hybrid, or remote

What I'm looking for

  • Lead or Staff product design roles
  • Developer tools, AI products, design systems
  • Austin, remote, or hybrid