AI will soon touch every aspect of the food industry, from agriculture, to logistics, to recipe development, and beyond. It’s already powering restaurant drive-thrus with AI ordering conversations; by writing marketing copy at marketing and media agencies; and by helping culinary teams envision new recipes from trends spotted in big data. It’s no longer a matter of if our existing systems and business models will be disrupted: we’re due for radical transformations far sooner than most of us would have predicted.
We’ve been building for this disruption since day one at Galley. Integrating AI into the kitchen operation has always been one of our goals, and it has informed how we’ve architected Galley’s data capabilities. It’s how we gained an AI-first fund as our first venture partner several years ago. Now that transformative and accessible AI is here, I want to explore how our strategy is playing out, and where we see this all going in a foodservice context.
The Building Blocks of Culinary Grammar
In 2017, Google’s Jakob Uszkoreit published an industry-shattering blog revealing a new framework for building artificial intelligence: “transformers”. Before transformers, there were many kinds of AI technologies and computational frameworks, and they were all advancing slowly, and independently. But the transformer model brought it all together under one idea: if you can condense a mode of information (text, images, video, brain scans, recipes) into language, you can train a model to understand the language, then create new insights back in the original mode, or even different modes (like turning text into digital art, or fMRI brain scans into video).
The transformer framework enabled all AI and machine learning disciplines to unify under one operating system, so to speak: the large language model (LLM). Now, any type of idea or data or pixel can conceivably be synthesized into language, enabling humans and computers to engage with it through AI interfaces. The language is the central mechanism for how AI learns and interacts with the world.
There is a grammar, of sorts, to it all. LLMs learn to spot patterns, rules, and norms in data, then turn that into language. With enough training, an LLM can understand the fine points of proper grammar, and even when it’s worth breaking those rules. It can learn the difference between the formulaic prose of a QSR menu, and how those rules are twisted and tweaked to form the poetry of fine dining.
At Galley, building the data infrastructure for food and culinary operations so that an LLM can learn to do this for the foodservice world has always been the vision.
To get there, an LLM must be able to learn the grammar of foodservice.
- How ingredients are combined to form recipes that taste good to people
- The relationship between inventory, demand planning, and purchase orders
- How to look at vendors and shave $0.13 off a plate cost to hit the right margin
ChatGPT can pretend to understand some of these relationships now, but push the model and you’ll quickly encounter unappetizing recipes, incorrect math, and recommendations so general they are useless.
Our industry needs an LLM that understands the nuances of foodservice, and creating the data architecture that enables an LLM to learn it is the only way to achieve that. That’s what we’re doing. And equipped with food and operational data spanning our entire industry, a Galley-powered AI will make for an incredible AI-powered Galley.
Two Ways AI Will Make Your Food Data More Productive
There are seemingly endless use cases for AI within food, but these are two avenues we foresee having a substantial and positive impact on the foodservice world.
Surfacing Deeper Insights through Algorithms
AI will eventually become the driving force behind Galley's internal algorithms and decision engine. Our holistic approach—considering not just recipes, but planning, inventory, production, and procurement—allows us to apply AI in a way that can find opportunities for efficiency gain that would be too difficult for a human to spot or know how to resolve.
Think about how challenging it is to create supply and demand parity with inventory. It’s always a game of cat and mouse: one week you over-purchase, the next you under-purchase, and back and forth you go—always aiming for perfect alignments, but never quite achieving it because the target is always moving.
Galley already uses complex algorithms to help food companies close the gap between food needed and food purchased, and this has yielded huge savings for our customers. But Galley’s algorithms optimize purchasing by looking at a defined set of variables that we’ve integrated into the algorithms. There are a thousand variables that have virtually no impact alone, but altogether, lead to significant impact on supply and demand. Accounting for all of these within our existing algorithms is not feasible. AI, however, will be able to learn how these microscopic variables work, and how to account for them in purchasing recommendations, without requiring
Chat-Based Interfaces for Kitchen Workers
The chatbots of the 2010s were inaccurate and endlessly frustrating. The chatbots of the 2020s can understand nuance, interpret unclear language like how humans do, and have reasoning engines that are impressive. The gap between these two technologies is wild.
It only takes three minutes of talking to ChatGPT to see that we’re already a lot closer to Captain Picard on USS Enterprise asking “Computer, tell me the nutritional makeup of turkey breast” than we are to trying to use a Facebook Messenger chatbot to check the store hours of a local restaurant in 2012.
In fact, we think that chat interfaces for kitchen workers will be one of the most compelling LLM applications within the food services industry. Imagine a bustling kitchen where chefs, instead of juggling screens and pans, can communicate directly with our system using conversational commands.
- “I’m tossing twelve spoiled tomatoes in the garbage” automatically updates your real-time inventory, as well as the actual cost of goods for the night
- “I need to double a recipe for spinach casserole” automatically pulls up the logged recipe for spinach casserole, along with the scaled units and updated production steps
- “How can I make a vegan version of our chicken tacos for a customer?” automatically pulls up an alternative recipe using in-stock ingredients that could be used as substitutes in the taco recipe, according to the nutritional metadata attached to those ingredients
Dashing between screens, pans, and knives is messy. Chat-based interfaces with recipes and production steps will make it easy for kitchen workers to input data into Galley, ask questions about procedures, or adapt to change in the kitchen in real-time.
It’s worth noting that neither of these are fully automated, AI-only scenarios. We firmly believe it’s not a question of AI versus humans, but AI with humans. The magic of human participation and intuition will always remain an essential part of the food experience, and hybrid working models will outperform those that rely nearly exclusively on either machines or humans.
In the hands of AI, our culinary operating system evolves into a dynamic tool that not only responds to users' commands, but also offers new levels of insights, optimizes workflows, and aids decision-making processes. The core data and system architecture, as well as the core workflows are what we've always been building towards at Galley, and are the reason we are poised to embrace and enable an AI-forward kitchen for our customers.
What’s a Rich Text element?
The rich text element allows you to create and format headings, paragraphs, blockquotes, images, and video all in one place instead of having to add and format them individually. Just double-click and easily create content.
Static and dynamic content editing
A rich text element can be used with static or dynamic content. For static content, just drop it into any page and begin editing. For dynamic content, add a rich text field to any collection and then connect a rich text element to that field in the settings panel. Voila!
How to customize formatting for each rich text
Headings, paragraphs, blockquotes, figures, images, and figure captions can all be styled after a class is added to the rich text element using the "When inside of" nested selector system.