Decode NDIS

AI-powered assistant that simplifies NDIS claims, funding, and policies for Australians with disabilities and their support teams.

Comment

About the Project

Decode NDIS is an AI-powered assistant that helps Australians with disabilities, plan managers, and service providers navigate the complex NDIS system - decoding invoices, understanding policy, and planning budgets in seconds.

What Inspired Us

Having worked at a plan manager in the past, we were inspired by the real struggles faced by NDIS participants who often face overwhelming bureaucracy. Australia's NDIS is generous in support but notoriously difficult to navigate. The average plan allocates $82,500, yet only $66,200 is spent, meaning essential services go unused.

How We Built It

Decode NDIS is built on a full-stack architecture with a clean, responsive UI and powerful backend logic:

  • Frontend: Built in Next.js with TypeScript, focusing on accessibility and responsiveness for a diverse user base.
  • Backend: Powered by Python Flask, responsible for routing queries and formatting responses.

  • AI Layer: We leveraged Perplexity’s Sonar API, strategically choosing the best model for each task:

    • sonar → factual code lookups
    • sonar-reasoning → policy logic
    • sonar-pro → real-time NDIS updates
    • sonar-reasoning-pro → budget planning with chain-of-thought

  • Data is enhanced by a curated NDIS support catalogue, including over 1000 official support codes.

Challenges We Faced

  • Domain complexity: NDIS rules are legally dense and frequently changing. Translating this into natural queries while ensuring accurate, compliant results was a key challenge.
  • Prompt engineering: We had to tune and route questions to the most appropriate Sonar model without overwhelming the user.
  • UX design for accessibility: Our users include individuals with cognitive and physical disabilities — this required thoughtful UI/UX planning, plain-language outputs, and mobile responsiveness.

What We Learned

  • How to map real-world policy logic into structured AI workflows
  • How to effectively orchestrate multiple AI models depending on the task
  • The importance of blending frontend with backend especially in accessibility-focused applications

Built With

Share this project:

Updates