How Cevo and AWS helped ECH launch a Claude-powered AI assistant for aged care onboarding in 10 weeks 

ECH partnered with Cevo to design and build ECH AI Assist, an AI-powered conversational agent that helps prospective clients and families navigate the early stages of the aged care journey. 

Using Cevo’s Knowledge Worker Accelerator, the team applied reusable patterns for conversational AI, knowledge retrieval, secure deployment and governance. Cevo’s engineers built the solution using Claude Code via Amazon Bedrock, with Amazon Bedrock AgentCore providing the agent runtime, identity and tooling needed to operate the assistant securely at scale. Bringing AWS and Claude together in this way let Cevo move fast on the engineering while keeping the solution inside ECH’s existing AWS environment, and freed the team to focus its own effort on the security and governance guardrails needed for a production AI system in aged care. This helped move the solution from idea to production in 10 weeks. 

ECH AI Assist is embedded into ECH’s website, giving users a simple way to ask questions about services, pricing, eligibility and My Aged Care registration. It captures enquiries and passes relevant context to ECH’s onboarding team via their CRM, helping them follow up with a clearer understanding of what each person needs. 

Cevo designed the assistant around real aged care onboarding journeys. Users can check whether ECH services their suburb, explore available services, understand funding pathways and get guidance on next steps. The assistant remembers key details during a conversation, such as suburb, funding program, pension status and services of interest, so users are not asked to repeat information. 

Supporting more complex aged care journeys 

ECH AI Assist supports Commonwealth Home Support Program, Home Care Packages Program, Support at Home program and private-pay pathways. It also reflects current aged care language, including the Single Assessment System and Support at Home program rules, while recognising legacy terms such as ACAT. 

For people receiving, or moving from, Home Care Packages Program support, the assistant can guide “no-worse-off” scenarios and part-pension contribution bands through a simple “calculate the dollar range” experience. For self-funded clients, it provides a dedicated pathway using ECH’s private price list. 

To make the experience easier for older Australians and family carers, the assistant uses quick-reply buttons throughout the journey. Users can move through funding choices, pension bands and closing options with less typing. 

Designing for accuracy, trust and safety 

Because aged care information and pricing can be complex, Cevo designed ECH AI Assist to ground responses in ECH’s approved content. The assistant draws on ECH service documents, pricing schedules and policy information before responding to factual questions. 

For pricing, Cevo used a deterministic calculator rather than relying on a large language model to generate dollar figures. This allows the assistant to calculate prices from ECH’s approved price list across different funding scenarios and reduces the risk of ungrounded pricing responses. 

Cevo also worked closely with ECH to put security and governance guardrails around the AI itself, not just the conversational experience. Additional safeguards were built into the experience: the assistant checks pricing responses before they reach the user, helps assess responses against ECH-approved policy and regulatory guidance, avoids medical advice, uses clear language and requests privacy consent before collecting personal details. 

A crisis-aware safety screen recognises emergency and domestic violence disclosures before AI processing and responds immediately with appropriate crisis contacts, including 000, Lifeline and 1800RESPECT. 

Building confidence for production 

Cevo delivered the solution end-to-end, covering the assistant experience, knowledge base, infrastructure, administration interface and deployment pipeline. Building with Claude Code via Amazon Bedrock, on top of Amazon Bedrock AgentCore, let Cevo’s developers move quickly through implementation, combining AWS’s infrastructure and Claude’s coding and reasoning capability to rapidly develop the solution for ECH and freeing the team to invest that time in the testing, security and governance work needed to bring an AI assistant into production safely. 

Quality assurance was built into the release process. Before every release, more than 40 scripted, multi-turn conversations are replayed against the live assistant, covering suburb checks, funding pathways, pricing quotes and safety scenarios. New behaviours are added to the test suite, helping ECH improve the assistant with greater confidence over time. 

ECH can also run adversarial red-team testing from the admin portal, with approximately 1,900 attack variations and downloadable reporting. A prompt-attack shield was added to block attempts to extract the assistant’s instructions, blocking 100% of test probes with no false positives. 

Cevo also incorporated its Model Optimisation Agent into the workflow, routing each query to the optimal Claude model based on the complexity of the prompt. This helps maintain response quality while managing token costs. The platform also includes live spend dashboards, budgets, circuit-breakers and cost isolation, while prompt-caching engineering reduced the cost per conversation by almost 70%, based on production traffic. 

The administration interface gives ECH a practical way to manage the assistant over time. Teams can upload documents, review enquiries, track conversation history, manage feedback, monitor spend, run red-team checks, manage suburb reference data and approve pricing updates before they go live.