Clearer, private visa preparation
DIY Visa
A privacy-first workspace helping Indonesian travelers prepare truthful U.S. B1/B2 visa answers, check consistency, and practice interviews.
Outcome
An end-to-end bilingual preparation journey combines guided questions, autosaved drafts, review tools, and personalized mock interviews.
My role
Full-Stack Engineer / Product Builder
How I helped
- Built the bilingual DS-160 preparation and review journey.
- Connected document review, photo guidance, and text or voice interview practice.
- Designed consent, encryption, ownership checks, and retention boundaries.
Product
The workflow in context
Captured from the live DIY Visa application on September 20, 2026.

The solution
DIY Visa grew from firsthand experience with confusing DS-160 questions. Applicants work section by section, review their own answers, catch possible inconsistencies, and practice a short interview in English or Bahasa Indonesia.
What changed
The hackathon MVP supports multiple private preparation cases, conditional questions, five-second autosave, human-readable review, photo guidance, and three-question text or voice interviews. Supporting tools include consent-based document auditing and optional retained files.
Behind the solution
Next.js server actions and route handlers call typed capability services. Supabase handles authentication and PostgreSQL storage through Drizzle. The Vercel AI SDK connects to an OpenAI-compatible provider, while private Vercel Blob stores explicitly retained documents.
Technical decisions
- Encrypt sensitive identity and financial fields before storing them in PostgreSQL.
- Enforce case ownership and tenant isolation on the server.
- Require consent for AI processing and confirmation before extracted facts update an application.
- Keep uploads transient by default and make retained-file deletion explicit.
- Keep model output schema-validated and expose uncertainty rather than approval predictions.
Key features
- Guided DS-160 preparation with conditional questions and autosave.
- Missing-answer, timeline, typo, and consistency review.
- Photo guidance and document review with traveler confirmation.
- Bilingual text or voice interview practice.
- Private document organization and a rolling retention policy.
Lessons learned
Trust needs to exist in storage rules, permission checks, interface copy, and user control. A focused preparation tool is more useful than trying to automate the entire government process.
Product boundaries
DIY Visa supports preparation only. It does not submit the DS-160, pay fees, book appointments, provide legal advice, or predict approval. Official applications remain under the traveler’s control on authorized government channels. The product is independent of the U.S. government.
AI system implementation
Typed server-side capability services use the Vercel AI SDK with an OpenAI-compatible provider for answer review, consent-based document auditing, and bilingual interview practice. Model output is schema-validated; extracted facts require traveler confirmation before updating a case. Ownership checks and encrypted sensitive fields protect stored case data. AI supports preparation, not submission, legal advice, or approval prediction.
Could a similar approach help your business?
Tell me where work gets stuck. We can identify a practical first improvement and what success should look like.
Discuss your challenge