Decode NDIS

When my colleague and I started as data entry clerks at an NDIS plan manager in 2022, we never imagined that just a year later we’d be presenting at the NAB Arena, showcasing an AI solution that could transform the work we once did manually. That experience came full circle when our project won 2nd place at AI Hack 2023, for developing a tool that helps NDIS participants scan invoices and automatically interpret activity descriptions into the correct numerical NDIS codes for insurance claims. Below are some of the details of the project: - Utilises GPT-4 Vision to extract uploaded invoice image data - Works with both handwritten and typed invoices - Supports manual text input for item descriptions - Uses LangChain and queries on Supabase vector database of official NDIS documents - Leverages OpenAI embeddings for semantic search - Outputs most relevant codes and summarises rules using GPT-4 Heres an example of how it would work: For a "Psychologist Appointment", the system provides multiple applicable codes ordered by relevance, considering factors like participant age and plan category. This multi-option approach ensures accuracy when item descriptions lack sufficient detail.