The Journey of Project PA2523 - Building an NDIS Reportable Chatbot
- Tommy Dam
- Jun 26
- 5 min read
Introduction
Innovation in healthcare technology often begins through strategic partnerships between industry and academia. In early January 2025, Carelogix reached out to Western Sydney University (WSU) with a research & development project to tackle one of the most challenging aspects of NDIS (National Disability Insurance Scheme) compliance through their capstone program.
Professional Experience is Western Sydney University's premier undergraduate capstone subject for computing students. It provides the opportunity to collaborate with a team of final-year students on web, mobile app, or software-based projects across the semester. Students gain industry-ready experience while industry partners benefit from fresh ideas, technical skills, and real progress on their project goals.
In March 2025, Carelogix was allocated student resources, marking the beginning of what would become Project PA2523. Over the next 15 weeks, four talented students from WSU embarked on an ambitious journey to tackle one of the most time consuming aspects of NDIS compliance - determining what constitutes a reportable incident.
The Challenge: Navigating NDIS Reportable Incidents
The NDIS reportable scheme presents significant challenges for service providers. Key issues include:
Complex Regulations: Under complicated regulations, NDIS providers experience difficulty determining whether incidents are reportable or not
Inconsistent Reporting: Underreporting of significant incidents and overreporting of smaller incidents
Subjective Interpretation: Inconsistent decisions caused by subjective interpretation
Manual Processes: Manual procedures lack transparency, are time-consuming, and susceptible to error
These challenges formed the foundation for our innovative solution - an AI-powered chatbot we affectionately named "Care Bot."
The Solution: Care Bot
We developed an AI-powered chatbot designed to help NDIS providers classify incident reports as "Reportable" or "Not Reportable" under strict compliance guidelines. The system combines AI with a rule-based fallback, all securely hosted on AWS in Australia to guarantee data privacy compliance.
Team Structure and Weekly Coordination
Our success was built on a foundation of clear role definition and consistent communication. Each team member took ownership of a critical component:
Cloud Developer: Managed AWS infrastructure and deployment
Frontend Developer: Built the React-based user interface
Backend Developer: Developed the FastAPI backend and integrations
AI Model Developer: Fine-tuned and optimised the classification models
Every week, without fail, we gathered after 5 PM for our project check-ins. These sessions became the heartbeat of our project, allowing us to address challenges, celebrate progress, and maintain alignment across all technical streams.
Weekly Progress Reviews: Building Professional Discipline
One of the most valuable aspects of Project PA2523 was our structured weekly progress review process. Every week, without fail, our team gathered after 5 PM for comprehensive status updates that went far beyond simple check-ins.
Meeting Structure and Objectives
Our weekly meetings were designed with clear objectives:
Progress Demonstration: Each student presented their work completed during the week
Barrier Identification: Open discussion of technical challenges, resource constraints, and blockers
Issue Resolution: Collaborative problem-solving sessions to overcome obstacles
Testing and Quality Assurance: Review of testing results, bug reports, and system performance
Continuous Improvement: Process refinement and optimisation discussions
Project Management: Timeline adjustments, milestone tracking, and resource allocation
Professional Development Through Accountability
These sessions provided invaluable professional development opportunities for the students:
Individual Accountability: Each team member was expected to demonstrate tangible progress within their designated role - whether cloud infrastructure improvements, frontend enhancements, backend optimisations, or AI model refinements. This created a culture of ownership and personal responsibility.
Team Collaboration: When individual challenges arose, the entire team rallied to provide support. We witnessed frontend developers offering insights into backend integration challenges, and AI engineers helping troubleshoot cloud deployment issues.
Communication Skills: Students developed the ability to articulate complex technical concepts, present their work professionally, and engage in constructive technical discussions with both peers and industry mentors.
Problem-Solving Approach: Rather than simply reporting problems, students learned to come prepared with potential solutions, alternative approaches, and risk assessments.
Typical Weekly Challenges and Solutions
Throughout the 15-week project, our meetings addressed various challenges:
Weeks 1-3: Initial setup challenges with AWS infrastructure and development environment configuration
Weeks 4-6: Model integration complexities and API endpoint standardisation
Weeks 7-9: Frontend-backend integration issues and user experience optimisation
Weeks 10-12: Performance optimisation, security implementation, and testing protocols
Weeks 13-15: Final integration testing, documentation completion, and deployment preparation
Each challenge became a learning opportunity, with students developing resilience, adaptability, and professional problem-solving skills that extend far beyond technical capabilities. In addition, each meeting also came with its own challenges ranging from camera hiccups to voice problems.
Impact on Project Success
This disciplined approach to weekly reviews was instrumental in maintaining project momentum and ensuring successful delivery. The meetings fostered transparency, encouraged proactive communication, and created a supportive environment where challenges were addressed promptly rather than escalating into major roadblocks.
The students' growth in professional presentation skills, technical communication, and collaborative problem-solving was evident week by week, preparing them not just for their final project demonstration, but for their future careers in technology.
Technical Architecture: A Modern Approach
Our solution leveraged cutting-edge technology across multiple layers:
Frontend (React Dashboard)
Built with Vite and React using modular CSS design
Responsive interface supporting desktop, tablet, and mobile
Intuitive chat interface with drag-and-drop file uploads
Real-time feedback and user experience optimisation
Backend (FastAPI)
Standardised API endpoints for seamless integration
PyMuPDF and Tesseract OCR for multi-format file parsing
Rate limiting via SlowAPI to prevent abuse
Comprehensive request validation and error handling
AI Classification Service
Custom-trained AI system using LLaMA 3.2 3B + LoRA
89% classification accuracy on test datasets
Structured JSON output with NDIS clause references
Performance optimised for sub-5-second response times
Cloud Infrastructure (AWS EC2)
Australian-hosted servers ensuring data sovereignty
HTTPS routing with security groups and access controls
Scalable architecture supporting future growth
Comprehensive logging and monitoring systems
Model Evolution: From Prototype to Production
Our AI development followed a methodical three-stage evolution:
Stage 1: GPT-2 Baseline
Initial proof of concept with 124M parameters
CPU-only inference with 60-65% accuracy
Binary keyword matching with limited contextual understanding
Stage 2: Gemma 3 Intermediate
Upgraded to 1.6B parameters hosted on Google Vertex AI
Improved accuracy to 75-80% with structured output templates
Enhanced reasoning capabilities with NDIS Section 8.2 compliance
Stage 3: LLaMA 3.2 Production
Fine-tuned 3B parameter model with LoRA optimisation
Achieved 89% classification accuracy using NVIDIA A100 hardware
Trained on over 200 synthetic and real examples with 160/40 train/test split
The Presentation Day
The culmination of our 15-week journey was marked by a formal presentation to an audience of university staff and other project clients. This capstone presentation represented the students' opportunity to showcase not just their technical achievements, but also their professional growth throughout the project. Standing before an audience of WSU faculty, industry representatives, and fellow students, our four team members demonstrated remarkable poise and professionalism. Each student presented their specialized contribution:
The presentation was more than a technical demonstration – it was a testament to their ability to communicate complex concepts to diverse audiences, handle questions under pressure, and articulate the real-world impact of their work. The presentation validated not only the technical success of the NDIS Reportable Chatbot but also the educational model that WSU's Professional Experience program represents.
Project Celebration and Future Horizons
After 15 intensive weeks of development, testing, and refinement, our team gathered in Parramatta for a well-deserved celebration dinner. The evening was filled with reflections on our journey, from those initial brainstorming sessions to seeing our AI successfully classify complex NDIS scenarios.
What made the celebration even more meaningful was the enthusiasm expressed by all four team members about continuing their journey with Carelogix. Their interest in pursuing internships within our startup operations speaks volumes about both their dedication to the project and their belief in its potential impact.
For any startup interested in getting support from WSU Professional Experience, please reach out to them, otherwise reach out to your local University for support, or leverage your network to keep moving forward.
NDIS service provider interested in the product should reach out to Carelogix directly to get exclusive access to use the chat bot.
Reference
Comentários