Employer platform · 2026

CaterVox — From catering conversations to quotes

Problem: A catering conversation needs the correct menu and policies, a saved request, and a clear handoff to restaurant staff.
My role: Full-stack and AI integration engineer at Crave, contributing location setup, restaurant context, estimate checks and staff workflows.
What changed: Connected published menu and policy context to the assistant and checked estimates against the relevant menu release.
What was verified: Saved October 2026 sandbox QA covers browser conversation, inquiry, staff quote and test payments; live phone/SMS rollout is not established.

CaterVox is Crave's catering voice assistant. I connected it to restaurant menus and catering policies, built location setup, and added estimate checks against published menu information. Saved sandbox QA follows a browser conversation through a persisted inquiry, staff quote and test payments. This case covers my integration work within the team, not ownership of the entire product.

01 / 04 · System overview

From a conversation to staff review

Catering conversation connected to restaurant menus and policies, a saved inquiry, and a staff-reviewed quote — CaterVox is Crave's product. This sanitized diagram shows the browser sandbox workflow; restaurant staff review the inquiry and prepare the quote.

01 Scope

The product context behind the work.

Problem

Catering requests involve more than answering a menu question. Guest counts, quantities, pickup details and restaurant policies must become information staff can review. The assistant also needs to distinguish an estimate from a staff-issued quote or a confirmed booking.

My contribution

Connected published restaurant menus and saved catering policies to the voice assistant's context.
Built assisted location setup and carried organization defaults into saved agent configuration.
Added estimate checks against published selling units and tied estimates and proposals to menu releases.
Worked on merchant settings, scoped customer context and staff-facing catering workflows.

02 Approach

Important engineering decisions.

CaterVox brings voice-assisted intake into a restaurant's catering workflow. My work at Crave focused on the connections between restaurant setup, the assistant's information and the staff tools that receive the request.

Use the restaurant's published context

I connected saved menu facts and catering policies to the assistant, alongside assisted location setup and organization defaults. That gives each location a defined configuration rather than relying on one generic prompt for every restaurant.

Estimates need their own checks. I added validation against published selling units and connected estimate and proposal context to menu releases. When menu information changes, the workflow calls for reconfirmation instead of silently reusing an old commercial context.

Keep staff responsible for the quote

The assistant can gather a request and provide an estimate, but an estimate is not a booking. The sandbox workflow saves an inquiry for staff to review, adjust and quote. This handoff keeps the conversation connected to the restaurant's operational records and human decisions.

My contribution covers these integration and workflow surfaces. The broader voice runtime, extraction system, commerce backend and product were developed by the Crave team.

Follow the saved sandbox result through the workflow

Saved QA from October 2026 records a human browser conversation becoming a persisted inquiry. The same sandbox journey continued through a staff quote, customer acceptance, test deposit and balance payments, and completion.

That result supports a bounded end-to-end sandbox example. It does not prove live telephone or SMS rollout, commercial customer adoption, or every reminder, cancellation, refund and recovery path. The diagrams use generic workflow labels and contain no customer or restaurant records.

03 Evidence

How the sandbox workflow works.

02 / 04 · Product journey

A request becomes a saved inquiry

Browser test conversation moving through request details and a saved inquiry to staff review and a prepared quote — The sandbox conversation produces a saved inquiry. Staff review the details and prepare a quote; the assistant does not independently confirm a booking.

03 / 04 · System boundary

Keeping the estimate tied to the menu

Published restaurant menu and catering policies feeding checked estimate context for the assistant and staff — Menu and policy context support the conversation. Estimate checks use published selling units and menu releases; changed information needs reconfirmation.

04 / 04 · Delivery / verification

What the sandbox demonstrated

Saved inquiry, staff quote, customer acceptance and test payments followed through to completion — Dated sandbox QA recorded inquiry persistence, a staff quote, acceptance, test deposit and balance, and completion. It is not production telephone or payment-adoption evidence.

04 Delivered & verified

What the public proof supports.

Delivered scope

Implemented scope

Location setup, menu and policy context, estimate validation and supporting merchant workflows.

Sandbox outcome

Saved October 2026 QA follows a browser conversation through a persisted inquiry, staff quote, customer acceptance and test deposit/balance payments.

Evidence boundary

The sandbox result does not establish live telephone/SMS rollout, paying customer adoption or every cancellation, refund and recovery path.

Stack

TypeScriptNext.jsNode.jsPostgreSQLAPI integrationVoice tools

Reflections

A useful voice workflow leaves staff with durable, reviewable information. A convincing conversation alone is not a completed catering request.
Published menus, selling units and policy checks matter because a fluent estimate can still be commercially wrong.

Next step

Have a similar project?

If this looks close to the product work you need, book a short call and I will help scope the next step clearly.

Book an intro call Email me