Skip to content
CASE 02 / SKIPLEGAL · CLIENT CAPSTONE · SPRING 2025

Designing an AI immigration workspace people trust with their most sensitive documents

  • Client Project
  • AI Tools
  • Multi-Step Forms
01 Context

SkipLegal is a startup using AI to guide people through U.S. visa, green card, and citizenship applications.

For our capstone, we designed its core workflow from scratch: securely uploading sensitive immigration documents, with AI that autofills forms in real time and a chatbot for support.

The catch: one typo can cost months and thousands of dollars. Would users trust AI with that?

Immigration process flowchart
02 Research · 4 interviews · 112 surveys · 15 testers

Overwhelmed

The process is unclear and repetitive, even with a lawyer.

Show your work

Trust in AI depends on being able to verify it.

Prove it's safe

Without a clear privacy explanation, users won’t upload.
03 Key decisions
Workspace wireframe with annotations

Four panels, introduced one at a time

  1. 1 Documents — secure uploads, forms one click away
  2. 2 Form viewer — the real government form, autofilled live
  3. 3 AI assistant — explains each question in plain language
  4. 4 To-Do — fixes become tasks, so nothing is lost
WHY Round 1 testers didn't know where to click first with all four on screen, so a tutorial now reveals them in sequence.

Eligibility quiz first

Problem → terms like “petitioner” made testers hesitate.
Decision → one question per screen + inline definitions; answers tailor the workspace.
Outcome → smoother onboarding in round 2.

Autofill or Manual

Problem → testers couldn't tell if the AI fills the form or just answers questions.
Decision → explicit mode toggle; AI explains what it fills first.
Outcome → testers felt more in control of their data, and overall trust rose from 3.0 to 4.1 out of 5 by round 2.

04 Edge cases & states
IF THE USER…THE UI…BECAUSE…
sees a legal term defines it on hover jargon caused hesitation
is new to the workspace reveals panels one by one full layout overwhelmed
is about to upload explains how data is protected privacy was the top hesitation
doesn't trust the AI offers Manual mode, one field in focus control builds trust
picks Autofill explains what's filled before acting no surprises with PII
makes a likely error flags it + adds a To-Do errors cost months
05 Before → after
Before: round-1 wireframe BEFORE

Round 1

Upload, form, to-do and chatbot all competed at once. Testers didn't know where to start, or whether the AI was filling the form.

After: final workspace AFTER

Final prototype

Landing page → eligibility quiz → workspace. Upload documents, watch the AI autofill the form, and ask the chatbot about any question.

IMPACT · ROUND 2 TESTING 3.0 → 4.1 trust rating out of 5
  • Usability — SUS rose from 69.7 to 78.2 (15 testers across both rounds)
  • Control — mode choice gave users agency over their data
  • Clarity — the eligibility quiz became easier to get through
  • Orientation — guided tutorial replaced first-view overload
  • Still open — some testers remained unsure of the chatbot's role

Final presentation

06 Reflection
Users of all backgrounds are capable of doing this. They know their histories, their dates, and documents. SkipLegal is just an AI tool to keep all of this information in order across dozens of forms, keeping the user in charge.

Challenges: as a startup, SkipLegal’s product and user requirements kept changing, so we re-scoped as its direction evolved. Shifting immigration policy made an already confusing process a moving target.

Next: have immigration lawyers test the flow to find gaps (and potentially endorse it), and build trust with testimonials and clearer security messaging.