uhcDOCS

Intake, consent, and billing ran on more than 40 paper forms. Instead of redrawing each one, we built the library that builds the forms, so the next one could be assembled in days instead of weeks.

  • RoleSoftware Engineer
  • CompanyUnitedHealth Group
  • Timeline~6 months
  • ToolsFigma, Visual Studio Code

*Some content has been redacted due to PII and confidentiality.

Intake, consent, and billing ran on 40+ paper forms.

Staff faced transcription errors, illegible handwriting, slow intake, and no audit trail when something went wrong.

My title was Software Engineer, but I owned the design layer.

On a team of 2 designers, 6 engineers, and a PM, I designed the components, defined the tokens, wrote the handoff documentation, and worked out workflows and edge cases. Sitting in both worlds let me close the gap where specs usually drift.

The audit: every intake, consent, and billing form, with the high-volume few as our beachhead

I started by auditing every form, down to each field.

I went through every field, signature line, and conditional branch. Support-ticket dashboards showed that a handful of high-volume forms caused most intake errors, so those became our beachhead.

Paper hides a lot of logic until you try to reproduce it.

Legal

Signatures with the same legal weight as paper

Our answer

An electronic signature flow

Compliance

A valid way for brokers to sign for others

Our answer

Broker attestations to complete and sign on behalf of employees and employers

IT

Something reusable, not 40 one-off builds

Our answer

Rule-driven components that assemble forms from configuration, not static layouts

Three groups held veto power, and each needed something different.

Legal, Compliance, and IT all had to say yes. I framed the goal narrowly enough for all three to agree: replace the highest-volume forms with accessible digital versions carrying the same legal weight.

Tokens

  • Color
  • Type
  • Spacing

Primitives

  • Button
  • Input
  • Alert
  • Progress

Form patterns

  • Multi-step
  • Conditional
  • Review-and-submit

Forms

Assembled from configuration

  • Intake
  • Consent
  • Billing

The bet: don’t design forms. Build the library that builds the forms.

We built it in layers: tokens first (color, type, spacing), then primitives (button, input, alert, progress), then form patterns (multi-step, conditional, review-and-submit), so the next form could be assembled in days instead of weeks. Every component held WCAG 2.1 AA, which is non-negotiable in healthcare intake. Bi-weekly usability sessions with product support admins, a design crit with engineering every sprint, and a full accessibility audit before each release kept it honest.

uhcDOCS: the signature page

Accessibility had to hold all the way to the signature.

Intake and e-signature flows have a few classic traps:

  • Accessible e-signature: a typed-name signature with an attestation checkbox, instead of a drawn one that excludes keyboard and screen-reader users.
  • Errors announced, not just colored: linked to their field, read out, and summarized on submit.
  • Focus management: moving focus to each new step's heading in multi-step forms.
  • Session timeout warnings that let people extend, so no one loses their work.
  • Conditional fields that announce themselves when an answer reveals new questions.
Halved
Form completion time
−78%
Intake errors
2 squads
Adopted the library

Forms got faster to finish and harder to get wrong.

Form completion time halved in the pilot, and intake errors dropped 78%. The documented library was adopted by two product squads, and a shared vocabulary between design, engineering, and clinical ops meant fewer one-off forms and faster handoff.

The tokens weren't the interesting part. Pairing them with real handoff documentation is what stopped spec drift, and that's what made the system spread beyond my team.

Selected screens

Landing Page
Landing Page
Landing Page 2
Landing Page 2
Form Page
Form Page
Contacts Page
Contacts Page
Signature Page
Signature Page

Let's talk.

A little about me: