WhatsApp AI Agent Developer — Clinic Booking Automation (LLM Tool-Calling + Practice Management API)

What we’re building An after-hours WhatsApp agent for a Singapore physiotherapy clinic. Patients message outside opening hours to book, reschedule, or cancel appointments. The agent handles the conversation and writes to the clinic’s practice management system (Plato Medical, a Singapore-native platform). Anything ambiguous or high-risk gets queued for staff review at open. This is a production system handling real patient appointments — not a demo. Scope • WhatsApp Cloud API integration (webhook, session handling, template messages for the 24h window) • LLM agent with tool calling for intent classification and entity extraction • Integration with the Plato Medical API for patient lookup, availability, booking, reschedule, cancellation • Deterministic guardrails around destructive operations — cancellations and reschedules must not depend on model judgment alone • Identity resolution logic (phone number is not a reliable patient key: shared family phones, partners messaging on someone’s behalf) • Handling of booking race conditions and idempotency • Escalation path to a human queue for anything the agent can’t resolve confidently • Conversation logging and a way to audit what the agent did overnight What we need from you • Demonstrated production work with WhatsApp Cloud API (Twilio experience is fine but Cloud API is preferred) • Real experience with LLM function/tool calling — Anthropic, OpenAI, or similar. Not intent-tree chatbot builders. • Experience integrating with a practice management or scheduling API. Plato specifically is unlikely — but Cliniko, Nookal, Halaxy, Jane, SimplePractice, or Calendly-class systems all count. Tell us which. • Sound judgment about failure modes. We care more about how you handle the slot that got taken between showing it and confirming it than about how naturally the bot chats. • Comfortable working against sparse API documentation Nice to have • Healthcare or allied health domain experience • Familiarity with Singapore/Malaysia context (timezones, phone formats, patient expectations) • PDPA awareness How to apply Skip the template. Open your proposal with the answer to this: A patient messages at 11pm to cancel an appointment. The phone number matches two patient records — a mother and her adult son, both with appointments this week. How do you handle it? Proposals that don’t answer this won’t be read. There’s no single right answer; we want to see how you think about it. Also include: one project where you integrated an LLM agent with a real booking or scheduling system, what broke, and what you did about it.