Voiced

Inspiration

Jordan studied 3 days straight and still failed his midterm. Not because he wasn't prepared — because nobody told him he was legally entitled to extra time. 28% of disabled students at four-year universities actually receive accommodations. 72% get nothing — not because they don't qualify, but because the system is too hard to navigate alone. No one built the tool to help them. We did.

What it does

Voiced is an AI-powered accessibility platform built specifically for Iowa State University students with disabilities. It takes a student from "I don't know if I qualify" to "my professors have been notified" in under two minutes. Students sign up, select their ISU classes, and answer one question — do you have medical documentation? If yes, they upload their doctor's letter or psychological evaluation. Voiced reads the document using GPT-4o Vision, extracts the diagnosis, cross-references it against ISU's official SAS documentation requirements, and automatically selects the correct accommodations. A fully formatted, ADA-compliant letter is generated and sent directly to ISU Student Accessibility Services in one click. For students without documentation, Voiced offers a voice intake path — they describe what they're struggling with and Voiced routes them either to the accommodation flow with guidance on what documentation ISU requires, or to real ISU support resources like Student Counseling Services and the Academic Success Center with real phone numbers and locations. The professor compliance tracker is what makes Voiced truly unique. After accommodations are approved, Voiced sends personalised notification emails to each professor via n8n automation. It tracks who has acknowledged the letter. If a professor hasn't responded in three days, Voiced automatically generates a follow-up email citing the exact federal law they are violating and sends it to their ISU inbox. No other tool does this.

How we built it

Frontend: React Native with Expo, running on both web and mobile from a single codebase across nine screens — login, class entry, diagnosis check, document upload, voice intake, roadmap, letter generation, professor dashboard, and professor portal. Backend: Node.js with Express and TypeScript. Seven REST API endpoints handle transcription, document analysis, letter generation, professor email generation, follow-up generation, resource generation, and direct email sending. AI pipeline: OpenAI Whisper converts voice recordings to text. GPT-4o analyzes transcripts and uploaded documents. Every GPT-4o call injects ISU's official SAS knowledge base into the system prompt — real documentation requirements, real accommodation criteria, real office details from sas.dso.iastate.edu. This is our RAG implementation. The AI doesn't guess. It reasons against ISU's actual published policies. Email automation: n8n cloud workflow. When a student clicks Send Letter or Follow Up, our backend calls the n8n webhook, which delivers the email to the professor's real ISU inbox via Gmail SMTP in under five seconds. Document processing: pdf-parse extracts text from uploaded PDFs. GPT-4o analyzes the text against ISU's disability-specific documentation requirements and returns structured JSON with the diagnosis, documentation sufficiency, applicable accommodations, and any gaps. Database: Supabase PostgreSQL for authentication and data persistence. Three user roles — student, professor, and admin — each with a tailored experience.

Challenges we ran into

Getting React Native to work reliably in the browser was our biggest technical challenge. React Native is designed for mobile — making it render correctly on web required solving ScrollView bugs, platform-specific layout issues, and Expo bundling conflicts with several packages that don't support web environments. The document analysis pipeline took significant iteration. GPT-4o Vision doesn't read PDFs directly — we needed pdf-parse to extract text first, then pass that text alongside ISU's complete knowledge base. Getting the prompt engineering right so the AI matched diagnoses to ISU's specific criteria, not generic accommodation advice, required many iterations. Handling fake and invalid medical reports was a real challenge we had to solve deliberately. Students could upload any document — a blank page, a receipt, a non-medical letter. We built detection logic into the AI prompt to identify whether the uploaded document contains legitimate medical or psychological diagnostic content. If it doesn't, Voiced tells the student clearly what ISU actually requires rather than generating incorrect accommodation recommendations. This was critical for ensuring the platform produces trustworthy, accurate outputs rather than hallucinated results. ISU's email system blocks third-party SMTP access. We had to architect around this using Gmail SMTP through n8n, which made the automation pipeline more robust and visually demonstrable during the demo. Coordinating a four-person team across frontend, backend, authentication, and database in 24 hours required clear API contracts from the start. We defined every endpoint before anyone wrote a line of code.

Accomplishments that we're proud of

We implemented a complete voice intake pipeline using OpenAI Whisper. Students tap the mic, speak naturally about what they're struggling with, and within seconds the speech is transcribed, analyzed against ISU's SAS knowledge base, and routed to the right support path — accommodations or ISU resources. Real voice. Real transcription. Real AI reasoning. Working end to end in the browser. We built real automated email delivery. Every Send Letter and Follow Up button sends an actual email to a real ISU inbox via our n8n automation pipeline. During our demo, judges can watch an email arrive on a teammate's phone in real time. This is not a mockup — it is a live working system. We implemented RAG — Retrieval Augmented Generation — grounded in ISU's actual published policies. Every AI response is based on real ISU SAS documentation requirements, real accommodation criteria, and real eligibility rules. The AI cannot hallucinate ISU policies because it is told exactly what they are before every response. The professor compliance tracker is genuinely novel. A semester-long dashboard that tracks which professors have acknowledged a student's accommodation letter and automatically generates ADA-citing follow-up emails when they don't respond. No tool like this exists anywhere. We built for genuine accessibility. Our colour palette meets WCAG AA contrast ratios. Our font stack uses OpenDyslexic with Verdana fallback. The platform for disabled students is itself accessible — which is more than can be said for most university accessibility portals.

What we learned

Real-time systems are harder than they look. Getting voice recording, Whisper transcription, GPT-4o analysis, and UI state updates to flow smoothly required careful error handling at every step of the pipeline. Prompt engineering is a genuine skill. The difference between a generic AI response and an ISU-specific legally-grounded response came entirely from how we structured the system prompt. Injecting ISU's actual documentation requirements transformed the AI from a chatbot into a domain expert. The most impactful features are not always the most technically complex. The professor compliance tracker took less time to build than the voice pipeline but delivers more unique value than anything else in the product. Solving the right problem matters more than solving a hard problem. Accessibility is a foundation, not a feature. Building a platform for disabled students with poor contrast, small touch targets, or hard-to-read fonts would have been a failure of the entire mission. We are glad we thought about this from day one.

What's next for Voiced

Expand to other universities. The ISU knowledge base is one file. Replacing it with any other university's SAS documentation makes Voiced work for that institution instantly. The long-term vision is a platform any university can onboard in hours. Canvas LMS integration. Instead of students manually entering their classes, Voiced connects to ISU's Canvas API and imports enrolled courses and professors automatically. Semester renewal reminders. Accommodation letters expire each semester. Voiced sends students a push notification two weeks before the semester starts reminding them to renew through the Accommodate portal. Microsoft Graph API for email. Instead of routing through Gmail, Voiced sends emails directly from the student's ISU address via the Microsoft Graph API — making every notification come from a verified ISU email. Research partnership with ISU SAS. We want to work directly with the Student Accessibility Services office to improve documentation matching accuracy and help more students get their applications right the first time.