Inspiration
An application form can be accurate yet difficult to use. People must interpret unfamiliar terms, identify required documents, and find the right information without losing track of the original instructions. We wanted to explore a more accessible experience where clarity and evidence appear together.
W3C guidance emphasizes clear labels, instructions, and accessible error handling. OECD work also stresses inclusive, human-centred public-service design. These principles informed our concept, but this prototype has not yet been validated in a live service.
What it does
SevaStep turns plain-text application instructions into a guided preparation flow. Users can inspect detected questions, answer them one at a time, check required information, and print a preparation summary. Every generated question links to its source line so users can verify the wording.
The demo uses a fictional Community Learning Support Pilot form. It does not determine eligibility, submit applications, or connect to a government agency.
How we built it
The prototype is a self-contained static web application using semantic HTML, responsive CSS, and JavaScript modules. A deterministic parser detects label-and-colon patterns, required-field markers, and common input types. Answers remain in browser memory. Validation runs locally, and user-provided text is safely rendered as text rather than executable HTML.
The interface supports English and Hindi navigation, larger text, high-contrast mode, browser text-to-speech, visible focus indicators, and keyboard-friendly controls. Imported text stays in its original language; machine translation is not implemented.
Challenges
Real documents are inconsistent. A colon may introduce a field, an instruction, or a warning. The parser excludes several non-field labels and preserves the original source text for manual checks, but it can still miss or misinterpret complex content. We used a fictional form to avoid encouraging reliance on unverified government rules.
What we learned
Traceability matters. A simplified question is easier to trust when users can see where it came from. The prototype also led to a product decision: a future AI model should suggest structures and explanations with confidence signals, while people retain control over corrections and final interpretation.
What's next
Potential next steps are PDF/OCR support; AI-assisted extraction with confidence thresholds; reviewed, multilingual explanations; source-document version tracking; and usability studies with applicants and frontline staff. We plan to evaluate interpretation accuracy, task completion, and accessibility. No user-impact or performance results are claimed yet.
Technology
HTML5, CSS3, JavaScript ES modules, Web Speech API, native browser printing, accessible form controls, and deterministic text parsing. This is a working proof of concept, not a deployed AI model.
Evidence and references
- W3C WCAG 2.2: Labels or Instructions
- W3C Forms Tutorial
- OECD Digital Government Outlook: Human-centred public services
Team
Individual team roles and contributions will be added after member confirmation and review. The team must meaningfully extend, test, and validate the prototype before final submission.
Built With
- accessibility
- css3
- html5
- javascript
- web-speech-api
