AI Use & Limitations

SRTD uses generative AI models (large language models, plus a separate model for voice transcription) to produce four kinds of output:

• Case note drafting. Given the participant, note type (STAR, SOAP, phone log, NDIA call log, progress, incident, behaviour, handover, etc.), and your rough input (typed or spoken), the AI produces a structured draft in language aligned with NDIS Practice Standards.
• Comms drafting. Provider emails, plan-manager queries, family updates, NDIA correspondence, replies to inbound messages, and texts - drafted in seconds. Where you have trained a writing-style profile, drafts adopt your voice.
• Document extraction. Given an NDIS plan, functional capacity assessment, behaviour support plan, discharge summary, OT report, psychology report, or plan-manager statement, the AI returns a structured summary: goals, funding, dates, recommendations, and follow-up tasks.
• Plan review analysis. Given the documents you have uploaded for a participant, the AI scores robustness across ten factors aligned to the NDIS 2026 rules and assembles a review-ready pack you can take into a plan meeting.

SRTD is a documentation and productivity tool, not an advisor or decision-maker. Specifically, SRTD does not:

• Provide clinical interpretation of the documents it analyses. A discharge summary or psychology report still needs to be read and interpreted by a qualified worker - the AI extraction is a starting point, not a substitute.
• Provide legal advice, including advice on NDIS legislation, the NDIS Code of Conduct, the NDIS Practice Standards, or any other regulatory framework.
• Provide financial advice, including advice on budget allocation or invoicing. Budget Tool outputs are estimates based on the documents you uploaded.
• Make decisions on your behalf. Every output is a draft for you to review, edit, and decide whether to act on.
• Advocate before the NDIA, the NDIS Quality and Safeguards Commission, the AAT, or any other body. SRTD is not an NDIS-registered provider and does not deliver NDIS supports.
• Guarantee any audit, plan review, or compliance outcome.SRTD's outputs are designed to support documentation aligned with NDIS Practice Standards; whether documentation meets compliance requirements ultimately depends on how it is used and verified by the worker.

Almost everything stays in Australia. Specifically:

• AI inference for case notes, comms, budget analysis, document extraction, and plan review runs in AWS Bedrock in the Sydney and Melbourne regions.
• Voice transcription is the one US-based exception: voice memos are sent to OpenAI Whisper for speech-to-text and returned to the Australian inference layer. The audio sent contains no participant identifiers. Voice input is always optional, and the disclosure appears next to the microphone button every time you use it.
• Application data (your account, participants, generated content, plan review sessions) is stored in Sydney-region data centres via Supabase.

For the full sub-processor list, see our Security & Privacy page.

Not your data.Your case notes, drafts, emails, voice memos, uploaded documents, and participant details are never used to train AI models - ours, Anthropic's, OpenAI's, or any other provider's.

We use zero-retention agreements where the underlying provider supports them (AWS Bedrock, Whisper). Where we don't control the dial, the commercial contract explicitly prohibits training on customer content.

The base AI models we use are trained by their respective providers (Anthropic, OpenAI) on a mixture of publicly available text and licensed content. We don't have visibility into the training mix beyond what those providers publish.

AI models can and do make mistakes. The most common failure modes you should watch for:

• Hallucination.The model can invent details that sound plausible but aren't in your input - specific dates, names, providers, or events that you didn't mention. Always check generated content against what actually happened.
• Outdated information. The base models we use have training cutoffs. References to NDIS policy, legislation, pricing, or guidance can be out of date. We refresh prompts and guidance as official documents are published - but always verify regulatory references against current sources before relying on them.
• Missed context.The AI only sees what you give it. If you didn't tell it about a behaviour support plan, a recent incident, or a specific provider relationship, the draft won't reflect that context. Review the output for what's missing, not just what's wrong.
• Tone mismatch. Even with writing-style profiles, the AI can occasionally misread the tone a situation calls for - particularly with sensitive comms. Always re-read before sending.
• Document misreading. Document Analysis can misclassify, mis-link figures across line items, or skip late-page content on long PDFs. The source document is authoritative.

Across every SRTD tool, the worker who uses the output - you - is responsible for what gets submitted, sent, or acted on. Specifically:

• Review every output before saving it to your practice management system, sending it as a message, or referencing it in any submission to the NDIS.
• Verify facts, figures, names, and dates against your source materials before relying on them.
• Edit anything that doesn't reflect what actually happenedor doesn't match your clinical, professional, or contextual judgement.
• Make the final decision on whether to use the output. SRTD is a draft assistant; you remain the decision-maker.

The in-app disclaimer under every AI output points back here. The same responsibility applies whether you used voice, typing, or a mix - and whether the output was produced in two seconds or after several regenerations.

If you notice an output that's seriously wrong, biased, or unsafe, please email support@srtd.tools with the subject line "AI output report". We use these reports to refine prompts, add guardrails, and improve the product - your feedback is the fastest way the next version gets better.

• Security & Privacy - data residency, sub-processors, retention.
• Acceptable Use Policy - what you can and can't do with SRTD.
• Terms of Service - the binding contract.
• Privacy Policy - what data we collect and how we handle it.