Description
Summary (EWHC) is a UAE-based home healthcare company offering at-home blood testing, doctor visits, IV therapy, and vitamin injections. We generate leads through TikTok, Meta, and Google Ads. We already use Wati for WhatsApp Business API, with a live Make.com scenario that routes form-based leads (from TikTok/Meta instant forms and Google Ads landing pages) into a Google Sheet and triggers a WhatsApp welcome template via Wati's API. Separately, we also receive Click-to-WhatsApp (CTWA) ad DMs — patients who click "Send Message" on our Meta/TikTok ads and land directly in our Wati Team Inbox as a live conversation, with no form and no Sheet involved. These currently rely entirely on a human being online to respond. The Project We want a custom-built AI agent — using an AI API directly (Claude or GPT-4, not a pre-packaged AI-agent) — connected to our Wati WhatsApp API via Make.com, so we control the logic and keep costs low instead of paying for a bundled AI-agent platform subscription. What the AI needs to do: 1-Respond to incoming WhatsApp messages from both flows — replies to our automated welcome messages, and direct CTWA DMs landing in the Wati inbox with no prior form/trigger 2-Hold natural conversations in Arabic (Gulf/Khaleeji dialect) and English, detecting which to use Answer questions accurately using our SOPs/knowledge base (service packages, what's included in each, general process) — must never guess or invent pricing or test contents 3-Guide the conversation toward booking an appointment 4-Once a patient is ready to book, collect: Name, Phone Number, Service/Package Requested, Preferred Appointment Date/Time 5-Send that captured data into Monday.com or Google Sheets in real time, so our human CS team can see it and make a confirmation call 6-Hand off cleanly to a human via the Wati Team Inbox for anything outside its scope (medical specifics, complaints, price negotiation, complex scheduling) — the AI should never be the final word on a confirmed booking 7-Sound like a person, not a scripted bot Proposed Architecture (open to your recommendations) 1-Wati webhook fires on every incoming message (both flows) into Make.com 2-Make.com calls the Claude or GPT API with the conversation + a structured knowledge base (SOPs, package data) for context 3-AI response is sent back to the patient via Wati's API 4-When booking details are captured, Make.com pushes a new item into Monday.com via its native module 5-Our existing Make.com scenario (form leads → Sheet → welcome template) stays as-is; this is a new, parallel scenario for the reply/conversation layer 6-Our Current Stack (must integrate with, not replace) 7-Wati (WhatsApp Business API), account active 8-Make.com for automation/orchestration (existing scenario already live) 9-Monday.com or Google Sheet as the destination for captured lead/booking data 10-Leads arrive via: TikTok/Meta instant forms, Google Ads landing pages, and CTWA ad DMs (direct to Wati inbox) Why Not Astra / Pre-Built AI Agent Platforms We evaluated Wati's own Astra AI agent and decided against it for cost reasons — we want a leaner, directly-billed AI API (Claude/GPT) instead of a $99+/month bundled platform subscription. Please build accordingly; we are not looking for a recommendation to use a packaged AI-agent product. Nice to Have: 1-Experience building AI agents with the Claude API or OpenAI API, connected via Make.com or similar 2-Experience with Wati specifically, or WhatsApp Business API more broadly 3-Experience connecting Make.com to Monday.com 4-Arabic-language conversational AI experience (Gulf dialect a plus) 5-Experience with Click-to-WhatsApp (CTWA) ad attribution/tracking 6-Healthcare, wellness, or appointment-booking business experience Deliverables 1-Live AI agent responding to incoming WhatsApp messages via Wati — covering both form-triggered replies and CTWA DMs 2-Make.com scenario(s) implementing the architecture above 3-Structured knowledge base (from our SOPs and package list) that we can update ourselves 4-Monday.com integration for captured lead/booking data 5-Clear escalation logic to human CS via the Wati Team Inbox 6-Short documentation/handover so our team can maintain it 7-Critical Requirement: Zero Hallucination on Prices & Test Lists This is the single most important requirement of this project. Our packages have similar but different tiers (99 AED, 139 AED, 299 AED, 499 AED, etc.), and the AI must never guess, blend, or invent a price or a test list. We expect the AI to be built using a retrieval-based approach (RAG) — meaning prices and package contents live in a structured source of truth, and the AI looks up the exact data for a package at the moment of reply rather than relying on what it "remembers" from a prompt. If a patient asks about something not in that data, the AI must say it will confirm and hand off to a human — it should never guess. To Apply, Please Include: 1-Relevant past projects (Wati, WhatsApp Business API, Claude/GPT API integrations via Make.com) 2-Your proposed approach for handling both lead flows (automated + CTWA DMs) within one system 3-A clear explanation of how you prevent the AI from hallucinating prices or test lists — we're specifically looking for a retrieval/lookup-based approach, not just "a detailed prompt." Please describe: Where pricing/package data would live and how the AI accesses it at reply-time What the AI does when asked about something not in your data How you test for hallucination before going live (e.g. adversarial/trick-question testing) 3-A past project where accuracy mattered (pricing, medical, financial, or legal information) 4-Your approach to connecting Make.com to Monday.com 5-Estimated timeline and price 6-Any limitations you'd flag with our current Wati/Make.com setup Send them as a "LOOM VIDEO" Thanks