Team Name

KoalaCoders

Team Members

Project Description

KoalAI Project Plan: A GovHack Initiative

1. Executive Summary

Project Name: KoalAI

Concept: A friendly, conversational AI assistant designed to help Australian entrepreneurs navigate the complexities of starting a business. KoalAI will act as a single, trustworthy source of information, integrating data from federal, state, and local governments to provide personalised, step-by-step guidance.

Governing Framework: This project will be developed in strict adherence to the Australian Government AI technical standard to ensure it is responsible, ethical, transparent, and safe.

Primary Use Case: Assisting a user with the life event of "Starting a Business."

!http://googleusercontent.com/image_generation_content/0

2. Guiding Principles (Whole of AI Lifecycle)

These principles will be embedded in every stage of the project:

• Explainability & Transparency: KoalAI will always clearly state it's an AI and cite the official government sources for the information it provides.
• Auditability: All interactions, data sources, and system versions will be logged to ensure full traceability and accountability.
• Bias Management: A comprehensive bias management plan will be implemented to actively identify, assess, and mitigate bias in data, algorithms, and testing.
• Human Oversight: A "human-in-the-loop" will always be an option. The system will be designed to escalate complex or high-risk queries to a human expert for review.

3. Phase 1: Discover (Design, Data, Train, Evaluate)

Design

• Problem Definition: Address the overwhelming complexity aspiring entrepreneurs face when navigating fragmented government regulations and information.
• Human-Centred Approach: The design will be co-developed with real entrepreneurs. It will be inclusive, meet WCAG accessibility standards, and include clear mechanisms for user feedback.
• Success Criteria: Success will be measured using a balanced set of metrics, including Value-Proposition (e.g., user-reported confidence), Performance (e.g., accuracy), Bias-Related (e.g., demographic parity), and Adoption (e.g., task completion rate).

Data

• Data Advantage: KoalAI's value comes from its ability to perform real-time data fusion and integration from multiple government sources, creating a personalised roadmap for the user.
• Data Supply Chain: The system will draw data from:
  • Federal: ATO, ASIC, business.gov.au
  • State/Territory: State-specific licenses and permits.
  • Local: Council zoning and regulations.
• Data Quality & Validation: We will use the ABS Data Quality Framework to ensure data is timely, accurate, and relevant. Automated validation checks will be implemented.
• Bias Management: We will ensure our datasets are representative of the diverse population of Australian entrepreneurs, actively managing for geographic, demographic, and industry-specific biases.

Train

• Training Mechanism: We will use a pre-trained Large Language Model (LLM). The core training technique will be Retrieval-Augmented Generation (RAG) to ground the model with our curated, trusted government data sources. This minimizes factual errors and allows for source citation.
• Tools:
  • LLM: An API-accessible model like Google Gemini.
  • Frameworks: LangChain or LlamaIndex to build the RAG application.
  • Vector Database: ChromaDB or FAISS to store and retrieve information.
• Training Environment: A secure cloud platform (e.g., GCP, AWS, Azure) with strict access controls, adhering to the ISM and Essential Eight.
• Model Validation: The RAG system will be validated for factual correctness, correct source attribution, and fairness across diverse user personas.

Evaluate

• Testing Strategy: Our strategy will include baseline testing (comparing AI answers to human experts), robust regression testing, and a high degree of automation.
• Mitigating Bias in Testing: The process will use independent testers, separate test data from training data, and a diverse group of test subjects for User Acceptance Testing (UAT).
• Safety and UAT: We will conduct adversarial "red team" testing to identify weaknesses and a final round of UAT with real entrepreneurs to validate the web app in real-world scenarios.

4. Phase 2: Operate (Integrate, Deploy, Monitor)

Integration & Deployment

• Security Approval: The project must obtain a Security Authority to Operate (SATO) before deployment.
• Phased Roll-out Strategy:
  1. Internal Pilot (Canary Release): Initial launch to a small group of internal government staff.
  2. Private Beta (Blue-Green): Controlled release to a limited group of real entrepreneurs.
  3. Public Launch (Traffic Shifting): Gradually increase the percentage of public users sent to KoalAI, from 10% to 100%.
• Safe Rollback: An automated rollback mechanism will be in place to instantly revert to the previous stable version if a critical failure is detected.

Monitoring

• Ongoing Monitoring: We will continuously monitor system health, infrastructure, and performance.
• AI Drift and Safety: We will actively monitor for AI drift (degradation in accuracy) and any unintended consequences or safety issues, using user feedback as a key input.

5. Phase 3: Retire (Decommission)

Decommissioning Plan

• Structured Process: The project will have a decommissioning plan from the start, outlining the scope, impact analysis, and a proactive communication strategy for users.
• Secure Shutdown: The plan will detail the technical steps to securely shut down all computing resources, wipe data, and terminate services.
• Final Reporting: All compliance records will be retained, and a final report detailing the process and lessons learned will be delivered to stakeholders.

Data Story

The Two-Year Detour: Ash's Dream and the Data Maze

Ash had a simple dream: to open a small coffee shop in their Queensland neighbourhood. The goal wasn't just to sell coffee; it was to build a community hub—a place for people to connect.

This dream was a small, local answer to a big national data story. [cite_start]Recent data shows that social cohesion in Australia has declined to its lowest point since 2007[cite: 41, 42]. Ash wanted to be part of the solution, one flat white at a time.

But then, Ash hit the data maze.

The journey started with a search for "how to start a business." [citestart]The data showed that for most people, this journey is "confusing and overwhelming, leaving them feeling powerless"[cite: 126]. [citestart]Ash discovered that key information was scattered across many different sources[cite: 125], from the ATO for an ABN, to ASIC for a business name, to the Queensland Government for a food license, to the local Brisbane council for a zoning permit.

Each click led to more complexity. [cite_start]Australia’s regulatory environment is a web of laws spread across local, state, and federal jurisdictions[cite: 56]. Ash spent months trying to understand their obligations, downloading different forms, and trying to figure out the right order of steps.

Ash's dream of building community was put on hold by a data problem. The passion that fueled the idea was being drained by bureaucratic friction. What should have been an exciting journey became a two-year detour filled with stress and uncertainty.

A New Data Story with KoalAI

This is the story KoalAI is designed to change.

[cite_start]By using AI to connect users with the right data from across government[cite: 2312, 2327], KoalAI transforms the maze into a single, clear path. It translates the complex regulatory landscape into a simple, personalised checklist.

With KoalAI, Ash could have completed their government requirements in weeks, not years.

By solving this data problem, we do more than just help one person open a coffee shop. We empower thousands of Australians to pursue their dreams, create local jobs, and build the community hubs that our data shows we desperately need. We can change the data story from one of frustration to one of empowerment.