How Spirit House Turned 30 Recipes Into an AI Cooking Assistant

Spirit House had already built something valuable: a collection of recipes that captured part of its culinary knowledge. The next question was how to make that information easier for customers to explore. In 2023, the hospitality business added about 30 recipes to Cody and began testing a conversational cooking assistant.

Evidence note: This case study reflects the owner’s reported experience in 2023. It does not assert continued use, current recipe volume, or results beyond the customer statements quoted below.

At a glance

• Organization: Spirit House, which describes itself as a restaurant and cooking school in Yandina, Queensland
• Challenge: help customers find and understand recipe knowledge through natural questions
• Approach: place about 30 recipes in a Cody knowledge base
• Reported response: the owner said people were enjoying the experience and shared an example involving a dietary question

From a recipe collection to a conversation

A traditional recipe page is designed to be read from top to bottom. A cook’s actual questions are rarely so linear. Someone may want to know which curry fits a preference, whether a recipe contains a particular ingredient, or which step can be prepared in advance. Those questions require the information inside the recipe—not simply a link to it.

Spirit House’s experiment was small and concrete. Rather than trying to encode every aspect of the business at once, it began with a bounded set of culinary content.

“Cody has about 30 recipes in the knowledge base, and people are loving it.”

Acland Brierty, owner of Spirit House

That structure gave customers a conversational entry point into material Spirit House had already created. The assistant could interpret a question, retrieve the relevant recipe context, and shape an answer around the user’s immediate need.

What an AI recipe assistant can help people ask

When the source material is complete, a recipe knowledge base can support focused questions such as:

• Which uploaded recipes use a particular ingredient?
• What order do the preparation steps follow?
• Which technique does the recipe call for at a specific stage?
• What substitutions or variations are explicitly documented?
• Which recipe passage supports the answer?

The important phrase is when the source material is complete. A conversational interface can make established knowledge easier to reach, but it cannot make missing or ambiguous information authoritative.

The green-curry question—and the lesson inside it

Brierty described one customer interaction that went beyond simple recipe retrieval:

“A guy wanted a recipe for a green curry and then asked if it was lactose-free, and obviously Cody knew that it was but that info was not in our knowledge base.”

The anecdote shows why conversational discovery feels useful: customers naturally ask follow-up questions. It also reveals an important limitation. If the lactose-free status was not present in the approved knowledge base, the answer may have relied on general model knowledge or inference. That should not be treated as verified dietary guidance.

For allergy, intolerance, or other safety-sensitive questions, a business needs a stricter standard. The answer should come from a maintained ingredient list, preparation notes, and cross-contamination policy. If that evidence is unavailable, the assistant should say so and direct the customer to a staff member. “I don’t have enough verified information” is better than a confident guess.

How to build a more reliable recipe knowledge base

1. Choose the source of truth. Use approved, current recipe files rather than copies scattered across documents and web pages.
2. Structure important details. Record ingredients, quantities, method, substitutions, dietary attributes, and known cross-contact risks consistently.
3. Separate culinary help from safety guidance. Allow broad cooking questions, but require source evidence and escalation for allergy-related answers.
4. Test real follow-ups. Include ambiguous requests, missing ingredients, local terminology, and questions the assistant should refuse to answer.
5. Assign an owner. Give someone responsibility for updating the knowledge base when a recipe or policy changes.

What Spirit House’s early experiment demonstrates

The supported result is qualitative, not numerical. Brierty reported positive customer interactions and enthusiasm for the direction of the product:

“I really like where Cody is heading and am happy to support it. We have some great stories about customer interactions with Cody.”

For other restaurants, cooking schools, and hospitality teams, the transferable idea is simple: valuable content becomes more useful when people can question it. The guardrail is equally important: answers should remain grounded in information the business is prepared to stand behind.

Give customers a better path into your expertise

Cody Assist lets teams organize business content, define how an assistant should respond, and make that knowledge available through a shared or embedded experience. Start with one well-maintained content set, then expand only after testing answer quality and escalation paths.

Explore Cody Assist to learn how to turn approved resources into a source-grounded AI knowledge assistant.

Organization context was checked against the official Spirit House website.