AI and automation / 2026
The capstone brief came from a fictional restaurant owner in Abuja: nobody answers the phone during busy hours, callers with simple questions hang up, and some do not come back. She wanted calls answered, orders and bookings confirmed back to the caller, and everything visible without anyone writing it down.
The voice agent runs on Vapi with the restaurant's menu, prices, hours and policies in its prompt. It has three tools, one each for a food order, a table reservation and a callback request, and each tool has a required schema. Every tool call posts to a single Make.com webhook with a request type. A router sends it to the matching Airtable table and returns a result to the agent, and a fallback route answers anything it does not recognise. A small dashboard reads the same records so staff can confirm a booking, start an order or return a call. I designed and built all of it myself.
Walkthrough
Built from the capstone recording: the Make.com scenario, a test call that reads the order back before logging it, the run log for that call, and the staff dashboard showing the order, a booking and a callback.
My part
Decisions
The prompt allows a tool call only after the assistant has the caller's name and phone number, the full details, and a spoken confirmation of the read-back. A half-finished order never reaches the kitchen's list.
A single webhook with a router keeps the tools pointed at one address and puts the routing rules in one place. Adding a request type means adding a route, not another scenario.
A request type the router does not recognise gets an explicit error response. The agent can tell the caller someone will follow up instead of assuming the record was saved.
The menu and delivery fees are written into the agent's instructions and it is told not to invent items, prices, areas or policies. In the recorded call it priced the order line by line and reached the correct total.
Screens
SYSTEM ARCHITECTURE
A caller reaches the Vapi assistant, which transcribes speech, generates replies with GPT-4.1 and speaks them.
When the caller confirms an order, booking or callback, the assistant calls the matching tool with structured arguments.
Each tool posts to one Make.com webhook. A router filters on the request type.
The matching route creates a record in its Airtable table and returns a success response to the assistant.
A fallback route catches unknown request types and returns an error response.
Built with
Work with me
If you are planning something similar, a 30-minute consultation covers feasibility, suitable tools and next steps. From $25.
More work
Gazelle's KitchenA WhatsApp business gatekeeper that qualifies genuine orders, enforces business and safety rules, contains off-topic or abusive conversations, and only surfaces the small set of cases that genuinely need the owner.PilotSellavie AIAn independent project: multi-tenant conversational-commerce software that connects social messaging with business knowledge, guarded AI, human handoff, orders, billing, analytics and operations.MaintainedAI Automation Control PlaneA runnable multi-tenant control plane for AI-assisted business workflows, combining FastAPI, Next.js, PostgreSQL, Redis, n8n, provider routing, human approval, idempotency, replay, cancellation, audit traces, and deterministic local demonstrations.Maintained