Constraining AI Agent responses on elder care pricing with ontologies and validated models
Understand this story
Based on practice in the Australian aged care industry, Safdar Ali shared a method in which an ontology supports a validated pricing prediction model, which an AI agent then calls to answer accommodation price questions, without fabricating any data throughout. In this industry, accommodation prices above A$789,686 require approval from a government agency and must be supported by evidence-based pricing justification; Ali says the relevant answers helped aged care providers meet this statutory evidentiary requirement. According to Ali, the model trained on more than 9,000 rooms across Australia can be completed using only a CPU, even though training includes neural network steps. However, when running the AI agent, providers may need to use third-party platforms that offer zero data retention (ZDR) options, such as AWS Bedrock, OpenRouter and Fireworks AI; for maximum data privacy, it can be run locally on mini PCs equipped with commercial-grade GPUs (such as Nvidia DGX Spark or Apple Mac Studio), which cost about A$9,500 and have low power consumption.
AI overview based on source reports · Updated 33 min ago
Coverage timeline
Follow the coverage to explore different perspectives.
- Hacker News AIOntology in the Age of AI
The author describes building an ontology and validated prediction model, orchestrated by an AI agent, to answer accommodation-pricing questions for Australia's aged care sector without fabricating information, noting the answers were said to help providers meet a legal requirement to evidence prices above A$789,686. The piece covers ontology definitions from Gruber and Noy & McGuinness, a worked retail example, the aged care pricing journey, and risks of black-box agents including data exposure and uncertain token costs. Key takeaways include that LLMs can be swapped but the ontology and validated model cannot, and that ontology should be designed before the agent is developed.
Activity for this story
Not enough continuous observations to show a trend yet.