Help a small-batch dessert company reduce food waste, preserve a centuries-old culinary craft, and train the next generation using applied AI!

This project supports UN SDG 12 (Responsible Consumption & Production) and SDG 4 (Quality Education) by reducing food waste, preserving culinary heritage, and training the next generation in sustainable, local, and all-natural culinary traditions using AI.

Gina’s Gelato is a small-batch dessert company in Nelson, BC, producing from-scratch gelato, sorbetto, and other desserts using all-natural ingredients and traditional food-processing methods. We avoid synthetic stabilizers, industrial emulsifiers, artificial colours and flavours, and lab-engineered additives, focusing instead on whole ingredients and small-batch techniques like blending, baking, steaming, cooking, and churning.

We are building an internal AI-powered tool to support real-time recipe development, ingredient substitutions, and product troubleshooting while adhering to our all-natural, from-scratch ethos ("slow food" philosophy). The tool will be trained using our in-house knowledge base, made up of our formulation PDFs, recipes, and handwritten or scanned notes. It must be expandable over time as more documents and data are added.

The goal is to create a system that allows our kitchen team—and future culinary students or interns—to create and/or adapt recipes, ask ingredient-related formulation questions, troubleshoot product issues, and understand ingredient impacts on texture, freezing point, stability, etc. without relying solely on the owner/operator (Gina), who currently handles all recipe creation, training, and technical oversight as well as other aspects of running the business. This tool will reduce food waste, improve kitchen efficiency, and serve as a reference and educational aid within our kitchen.

There are very few formal educational programs dedicated to from-scratch gelato or dessert formulation—especially those rooted in all-natural ingredients and traditional methods. This centuries-old craft is increasingly being replaced by lab-engineered bases and industrial shortcuts, even in artisan-labelled products. This project helps protect that knowledge in a small but meaningful way by building a tool that allows our team—and future culinary students or interns working with us—to learn directly from our recipes, scientific reasoning, and process logic. It supports the preservation of time-honoured techniques by helping users bypass modern artificiality and work confidently with whole ingredients, including local and/or organic when possible, while also streamlining the recipe development process for our core staff.

While the tool is designed to support recipe generation and learning, all proprietary recipes, PDFs, and internal documents used to train the system must remain hidden and inaccessible to users. These materials are to be used only for training the AI to provide accurate, science-informed outputs—without being exposed or downloadable.

Student collaborators will:

• Assess feasible technical approaches suitable for a non-technical business owner
• Design a user-friendly interface for querying gelato/sorbetto/frozen dessert-specific questions
• Recommend a scalable, low-maintenance, and cost-stable system
• Ensure the system allows uploads of new recipe scans, PDFs, or photos to expand the knowledge base over time
• Ensure users can generate complete, functionally correct recipes through queries, but cannot view, download, or access the underlying training materials (e.g., PDFs, scanned notes, proprietary recipes). These source documents must remain hidden and protected, used only to inform the AI’s responses—not exposed directly to the user.
• Ensure adherence to provided constraints (including our all-natural ingredient standards, from-scratch methods, slow food philosophy, and reference materials such as our PDFs and handwritten notes)
• Minimize AI hallucinations, as inaccurate outputs affect real-world business operations, job performance, confidence, and education.

Deliverables will include a research-backed platform recommendation, a working prototype (if feasible), and documentation for setup, usage, and future expansion.

• A research-informed system design, completed as part of the project, to determine the best approach for building a stable, cost-effective, and non-technical AI tool for internal business use
• A fully functioning, production-ready AI-powered tool with an intuitive front-end interface (similar to ChatGPT or Perplexity), designed for everyday use by kitchen staff and interns
• Complete setup and configuration (including hosting, if applicable), delivered turnkey so the system is ready for immediate use without requiring post-project technical development
• Role-based access controls that:
• Allow only the owner (Gina) or designated users to upload or manage training content
• Restrict access to the underlying training materials (e.g., PDFs, handwritten recipes, formulation notes), ensuring these documents are not visible, downloadable, or accessible to general users
• Limit download, copying, or screenshotting where feasible to protect proprietary knowledge
• Functionality to allow the owner to add new documents (PDFs, scanned recipes, photos) to the system over time without needing developer support
• Authentication is required for all access. Depending on the chosen platform, this may involve password protection, login-based access, or limiting access to specific IPs or users. Public-facing endpoints must be disabled unless explicitly authorized
• Clear, user-friendly documentation for using the tool, training new staff or interns, uploading new materials, and maintaining alignment with Gina’s all-natural ingredient standards and recipe style

Note:

This project must result in a fully functioning, production-ready tool for use in a real kitchen environment (not a demo or prototype). It should be stable, affordable, and easy to use without technical support, and all access to proprietary training materials must remain tightly controlled.

The tool must be private and secured from public access. It should only be accessible to the business owner and authorized team members. Public or unauthenticated users must not be able to view, query, or interact with the system in any way.