Title: AI Agent Engineer -Build Two Voice AI Agents (Customer + Service Provider) - marketplace APP

ABOUT US Hygea is a Melbourne-based home and commercial cleaning platform operating across Victoria, NSW and Queensland. We have live iOS and Android apps, plus an existing backend, pricing engine and booking system already running in production. We are now building our AI layer and need an experienced AI agent engineer to deliver it. THE PROJECT Two AI agents, both connected to our existing platform. Customer AI Acquisition and Conversion Agent Takes an existing lead through to a confirmed booking, and stops there. The agent calls the lead, holds a natural qualification conversation, collects property and service details, calls our existing pricing API for the quote, handles objections, and negotiates only within approved commercial rules. Once the customer agrees, the agent prepares the booking and sends them a secure link into our existing app, pre-filled with everything already collected. In the app the customer reviews the details, accepts the terms and legal documents, and taps one button to confirm the booking. That is the end of the agent's job. Everything after confirmation, including payment, service provider matching, scheduling, tracking, job completion, reviews, cancellations and rebooking, is already built and handled by our existing platform. None of those flows are being rebuilt or modified. Service Provider (SP) AI Agent Recruits and onboards cleaners into our business app. It contacts approved SP leads, guides sign-up, walks them through ID, public liability insurance and police check uploads, tracks verification and onboarding status through our own API, and routes verified SPs to human approval. Once active, it surfaces suitable jobs, supports authenticated job acceptance, encourages additional and recurring work, and explains approved reward milestones. Both agents follow the same fixed architecture principle: AI decides, a controlled tool and API layer executes, and our existing backend remains the source of truth. No direct database access. The freelancer will not be given access to our payment system. The AI agents cannot charge, capture, refund, cancel or process any payment. All payment, tracking, service and cancellation flows stay entirely inside our existing platform. WHAT YOU WILL BUILD The agent and orchestration layer. You recommend the approach, with rationale. A restricted allow-list tool layer over our existing APIs, with authentication, least-privilege scopes, schema validation, logging, timeouts and idempotency. Voice integration for outbound and inbound calls, with the core agent kept channel-independent so chat, SMS, WhatsApp and email can be added later without a rebuild. Commercial guardrails for the customer agent: maximum discount, price floor, eligible services, bundle rules, approval thresholds, offer expiry and a full audit trail, all admin-configurable rather than hard-coded. Verification and approval gates for the SP agent: document status, onboarding status, human-authorised approval, job matching, duplicate acceptance prevention and configurable reward rules. A secure, expiring handoff link that carries collected data into the app so nobody re-enters information they have already given us. An emergency stop that sits outside the model and prompt: master AI off, agent-level off, action-level off, pause, and human-only resume. Audit logging and correct attribution, duplicate prevention, explicit failure handling, and a QA plan covering unit, integration, end-to-end, security and emergency-stop tests. YOU MUST HAVE Production experience building AI agents that execute real actions through APIs. Shipped work, not prototypes. Strong LLM engineering: tool and function calling, orchestration, state and session management, prompt guardrails. Voice AI experience (LiveKit, Vapi, Retell, Twilio Voice, ElevenLabs or similar). Solid backend and API integration skills, including authentication, webhooks and idempotency. A security-first mindset. These agents speak to real customers and real contractors on our behalf. Clear written and spoken English, with overlap with Australian business hours (AEST). NICE TO HAVE PHP backend, Firebase and Twilio experience Workflow orchestration tools such as n8n Marketplace, booking or two-sided platform experience Australian client experience WHAT WE PROVIDE Documentation is complete for both agents: PRD with user stories and acceptance criteria, architecture and integration documents, technology requirements, and UI/UX designs. Shortlisted candidates receive these before quoting. Our in-house tech team owns the existing platform and will support the API work. HOW WE WILL RUN IT We will start with the Customer Agent as Milestone 1 and extend to the SP Agent once it is proven. Please quote both separately so we can plan the phasing. TO APPLY Please answer these directly. Generic proposals will not be reviewed. Describe one AI agent you have shipped to production that executed real actions through APIs. What did it do, and what broke that you had to fix? Which orchestration approach would you recommend here, and what are the trade-offs? How would you build an emergency stop that the AI cannot override or re-enable? How would you guarantee the agent never tells a customer a booking is confirmed when the API call actually failed? Your estimated hours for each agent, your hourly rate, and your earliest start date. Start your proposal with HYGEA AI so we know you have read this.