Proofly

Inspiration

As an international graduate, I experienced how immigration information is scattered across I-20s, I-94s, EAD cards, emails, award letters, and government websites. Even simple questions can require checking several documents and interpreting dates correctly.

I built Proofly to help immigrants understand what their documents say, what evidence they have, and what they should do next.

What it does

Proofly turns immigration documents into:

A dashboard with status, work authorization, and important dates.
A document-grounded chatbot with citations.
An O-1A evidence plan covering all eight criteria.
A prioritized list of missing evidence.
Immigration updates from official government sources.

Proofly evaluates document readiness, not visa eligibility, and marks information requiring professional review.

How we built it

Proofly uses React, TypeScript, and Vite for the frontend and FastAPI with Pydantic for the backend.

Supermemory processes documents and retrieves relevant evidence. DeepSeek-V3.2 through Featherless AI performs structured extraction, grounded question answering, and O-1 evidence analysis. Tavily retrieves policy updates restricted to validated government domains.

Every AI response is schema-validated. Citations must match documents supplied by the backend, and unsupported answers are rejected. If no relevant evidence is found, the chatbot refuses to invent an answer.

Challenges we ran into

The hardest challenge was keeping immigration concepts separate. An I-20 program date, EAD expiration, visa expiration, and authorized stay do not mean the same thing. Proofly uses separate models and never invents an expiration date for Duration of Status.

We also had to prevent the AI from making unsupported eligibility claims. Instead of displaying a misleading percentage, Proofly separates documented, partial, and missing evidence.

Because immigration documents contain sensitive information, the public demo is read-only and uses only fictional documents.

Accomplishments that we're proud of

Built and deployed a complete working application.
Created a grounded RAG chatbot with validated citations.
Evaluated all eight O-1A evidence criteria without making legal conclusions.
Correctly treated a future judging invitation as planned evidence, not completed judging.
Reached 201 passing backend tests and 29 passing frontend tests.

What we learned

Responsible AI requires more than a disclaimer. Safety must exist in the data models, retrieval process, validation rules, and refusal behavior.

We also learned that immigrants need transparent evidence and next steps—not another unexplained score.

What's next for Proofly

Next, Proofly will add authenticated user workspaces, encrypted storage, Google Drive import, persistent analysis history, deadline reminders, attorney collaboration, multilingual support, and additional immigration pathways such as EB-1A and H-1B.

Proofly does not replace an attorney. It helps immigrants become informed, organized, and prepared.