AIRest Receptionist

An eGain AI Agent that answers restaurant guests from the venue's own content and files table requests through a live API, with phone and email handover to the host team.

Comment

Inspiration

Restaurants lose bookings at the busiest hour. On a Friday at 7 PM the host is seating guests while the phone rings and website chats wait. We build AIRest, an AI receptionist for small restaurants, and wanted to see how far one eGain AI Agent can go: answer questions with sources and file a real booking request in the same conversation.

What it does

  • Answers guest questions (hours, location, menu, allergens and dietary options, reservation rules, parking, services) from the restaurant's own knowledge files, with the source shown under each answer.
  • A pre-chat workflow ("Guest Visit Details") collects the guest's name, party size and visit date before the chat starts, then offers to file a table request.
  • Saying yes, or typing "book a table", starts the process workflow "Table Booking Request", linked to the intent "Table Reservation". It skips questions the guest already answered, asks for time, email and special requests, and sends groups over eight to the events team.
  • An API action files the request with our booking service (Cloudflare Worker). The service checks opening hours and returns a reference number such as ME-M64H; the host team sees new requests on a /requests page.
  • When the content has no answer, the agent says so, offers the closest fact it has (for example the heated patio when asked about a rooftop terrace) and gives the host team's phone and email. A request for a person ("I would like to talk to a person") shows Call and Email buttons.
  • The demo runs on a clearly labelled fictional restaurant, Maple Ember Kitchen. Phone and email are placeholders.

Try it

Open the eGain demo link and click the chat bubble.

  1. Pre-chat: give a name, party size 2 and a date such as "Saturday, Nov 14".
  2. Click "Yes, book a table", then answer time (7:30 PM), any test email and a special request: the API action files the request and returns a reference number (ME-XXXX). Filed requests are listed at https://airest-egain-demo.ctmakc.workers.dev/requests.
  3. Restart and click "Not now, I have a question", then ask "Is there parking at the restaurant?" (answer with sources) or "Do you have a rooftop terrace?" (knowledge gap, host contacts).
  4. Type "I would like to talk to a person" for the Call and Email escalation. A party of 9 or more is routed to the events team.

How we built it

  • eGain AI Agent (trial) and Agentic Studio: knowledge from the restaurant's HTML content files, pre-chat workflow, process workflow tied to an intent, API action node, phone and email escalation, published experience and demo link.
  • Booking API and demo page: Cloudflare Worker with KV storage, deployed with Wrangler.
  • Every path was tested in Test AI Agent and through the public demo link on November 4 and re-tested through the demo link on November 6.
  • The 1-minute video was recorded with headless Playwright on the public demo link, with English captions burned in.

Challenges we ran into

  • The API node calls our server from eGain's own infrastructure. A {{variable}} in the request that no node sets sends the whole call to On Error without reaching the server, so every question got its own variable key.
  • Date and time variables could not be inserted into text fields, so date and time are asked as text questions and parsed on the server.
  • The pre-chat passes its history to the agent, which sometimes started the booking workflow on its own. A confirmation question at the start of the process workflow fixed it.

Accomplishments that we're proud of

One conversation both answers from the restaurant's content and files a booking request that a host can act on, with a reference number the guest sees in the chat.

What we learned

Most of the work in a support bot sits in the edges: what happens when a guest has already given details, when a group is too large, when the content has no answer. Workflows with clear handover rules matter as much as the answers.

What's next for AIRest Receptionist

Connect the booking action to real reservation systems, add the same flow to the phone channel of AIRest, and pilot it with a small restaurant.

Pre-existing templates, code and third-party assets

eGain AI Agent trial platform and its standard widget, workflow blocks and demo-link page. Demo knowledge files, booking API (Cloudflare Worker) and demo page written by the team. Food and interior photos on the demo page are AI-generated illustrations of the fictional restaurant. All contact details are placeholders (+1 555 number, example.com email).

Built With

Share this project:

Updates

Submission history