-
-
Supported procedure: original Chinese passage, source date and marked unofficial AI English translation.
-
Insufficient evidence: this guide cannot establish an individual reimbursement percentage.
-
Conditional procedure: retain regional, eligibility and document limits; an application is not approval.
The idea
Source Evidence Card explores a simple question: can a small document lookup stay useful while making its evidence and limits easy to inspect? It is intended for readers checking procedural information in one selected public Chinese health-insurance guide. This is an exploratory prototype, not a validated personal pain point or a medical-benefits decision service.
What it does
Enter an English or Chinese question. The app checks the downloaded source and returns an original passage with its section, publisher, publication date and official link. It preserves conditions when a procedure applies only in some circumstances. If the guide cannot establish a claim, such as a personal reimbursement percentage, it reports insufficient evidence. Two demonstrated passages include clearly marked, unofficial AI English translations.
How it was built
An AI coding agent prepared draft scope, product requirements, a technical specification and a build checklist, then implemented a Python standard-library server and a buildless HTML, CSS and JavaScript interface. The project uses the pinned Devpost Learn Skill Pack. Independent AI review found boundary, cache-validation and request-state issues that were corrected. The current implementation passed 40 automated checks, a fresh-copy run and desktop/mobile browser checks. These are development checks, not a user study or an accuracy benchmark.
The main challenge
A related passage is not necessarily evidence for a conclusion. The prototype uses inspectable procedural rules and conservative abstention, with limited topic coverage rather than unrestricted policy answers.
Authorship and current limits
This submission draft and the implementation were prepared with AI assistance at the participant's request. Participant plan review, hands-on validation and personal learning reflections are not established by automated checks. The project makes no claim that AI activity proves participant learning or real-world effectiveness.
Log in or sign up for Devpost to join the conversation.