Team Name:
V-SARC
Team Members:
V-SARC
The project developed an AI system, implemented as a chatbot, to reliably, transparently, and audibly assist in managing and interacting with government datasets. By combining Retrieval-Augmented Generation (RAG), vectorized metadata, and a tool-calling architecture, the system delivers grounded responses with full audit trails. Advanced AI techniques ensure ethical, accountable, and privacy-preserving outputs. The system enhances government decision-making with over 99% multi-factor confidence scoring.
Find the Evidence of Work here: Google Drive Folder
An Accurate and Trustworthy Chatbot for Data Interactions
Government agencies across Australia manage vast volumes of critical data. Despite this, converting these resources into actionable insights remains a major challenge. High-stakes decisions require near perfect accuracy, transparency, and accountability standards that traditional AI often cannot meet.
Figure 1: System Architecture Diagram
We designed an AI system focused on reliability, transparency, and inclusivity, combining advanced AI techniques with governance measures:
AusTender Procurement Statistics – Go to Dataset
Details of planned procurements, tenders, standing offers, and awarded contracts.
Freedom of Information Statistics – Go to Dataset
FOI activity reports including reviews, costs, and adjusted staff time.
Kaggle Employee Leave Tracking Data – Go to Dataset
Employee leave details for 2024 across departments.
Focus areas for strategic implementation:
Government agencies hold vast datasets but often struggle to extract insights. Traditional AI achieves ~90% accuracy, insufficient for critical decision-making.
Our RAG-based chatbot ensures:
- Grounded, scope-limited responses
- Human-guided interpretation
- Reduced hallucination risks
Tool calling enables AI to interact with external APIs and systems.
Figure 1: System Architecture Diagram
Steps:
1. Define the tool in the chat request.
2. Model decides to call the tool with parameters.
3. Application executes the tool.
4. Result processed by application.
5. Result returned to model.
6. Model generates final response with new context.
Figure 2: Tool Calling Overview
Reference: Spring AI Tools API
An AI chatbot that is accurate, trustworthy, auditable, and inclusive, designed specifically for government decision-making.
Government agencies oversee vast volumes of datasets but often struggle to extract meaningful insights efficiently. Traditional AI chatbots show potential but fail to meet the extremely high accuracy standards required in public sector decision-making, where even a 90% correct response rate may be insufficient. This gap creates challenges in trust, accurate, and accountability.
While recent advancements in AI, including large language models with advanced reasoning capabilities, promise sophisticated analysis, they also introduce risks such as hallucination, where outputs may be factually incorrect or unsupported by reliable sources. For government applications, accuracy and verifiability outweigh advanced reasoning; AI systems must provide grounded, scope-limited responses and allow human users to guide interpretation and decision-making.
Description of Use This dataset was used to generate statistical diagrams to demonstrate dynamic statistical diagram generation using LLM and dynamic tool calling.
Description of Use The dataset metadata was extracted to be used in the vector database. These metadata was used as a part of similarity search function PoC testing. This variety of datasets enables the dynamic nature of data fetching based on user prompts recieved.
Description of Use The data was scrapped and extracted to be used for a vector database. This vector database is simulated as a third-party data source for Retrieval Augmented Generation. The summarised statistics was therefore used to respond to user prompts with back references to the original data sources.
Go to Challenge | 12 teams have entered this challenge.
Go to Challenge | 18 teams have entered this challenge.
