User
Turing is designed for an R&D formulator, a specialist who designs and develops product formulations through research, experimentation, and testing to achieve desired product characteristics.
i.e. The lead formulator for Cookies Inc. is primarily responsible for developing and refining recipes that meet both consumer preferences and quality standards. She conducts experiments to adjust ingredient ratios, optimize baking conditions, and ensure consistency in flavor and texture.
Product
Items that are used daily by average consumers and require routine replenishment.
i.e. A chocolate chip cookie.
Product Category
Classification of products that share similar characteristics and serve similar needs or purposes.
i.e. A chocolate chip cookie could be classified as Snack.
Formulation
The recipe for a product.
i.e. The receipt (ingredients + their amounts) for a chocolate chip cookie.
Outcomes
The key parameters used to determine the quality of the formulation.
i.e. Target Hardness of the chocolate chip cookie is between 200-400 grams of force.
Outcomes Priority
The order in which the optimizer will attempt to find the solution, i.e. it will try to solve for the highest priorities first, then the medium ones and lastly the lowest priority ones.
Ingredients
The individual components or substances used to create a recipe or product
i.e. One recipe of chocolate chip cookies is made of Flour, Sugar, Butter, Eggs,, Baking soda, Salt, Vanilla extract and Chocolate chips.
Processing Conditions
The specific parameters and steps followed during the preparation and production of a recipe or product.
i.e. Ideal cooking temperature of a chocolate chip cookie is 325°F.
Inputs
A generic way to call both ingredients and processing conditions.
i.e. Inputs for a chocolate chip cookie could be Flour (ingredient) and Cooking Temperature (processing condition).
Descriptive Functional Role
The Descriptive Functional Role refers to the specific functions or purposes that each ingredient and processing condition serves in achieving the desired characteristics of a product. This field is key for the system to figure out the relationship between ingredients and outcomes in the Project Setup page. i.e. eggs serve as binding agents, adding structure and moisture while contributing to leavening and color.
Past Formulation Data
For product development, past formulation data refers to historical records of formulations, including the combinations of inputs (ingredients and processing conditions) and their associated outcomes. This data is critical for understanding patterns, identifying relationships, and building predictive models to guide future formulation decisions.
i.e. A dataset containing historical formulations of chocolate chip cookies—including variations in sugar content, baking temperature, and their resulting chewiness levels—can reveal insights such as the impact of sugar reduction on texture. This data forms the backbone for building models that predict and optimize future formulations.
Constraints
Specific product characteristics, represented as inputs, that the user enforces on the resulting recommendations.
i.e. In order to meet the marketing claims, sugar in the chocolate chip cookies cannot be more than 3 grams per serving.
Non-Negotiable Constraints
Non-negotiable constraints are absolute requirements that must be satisfied for a formulation to be considered valid. They differ from normal constraints by their strictness: while normal constraints can be optimized or adjusted to meet other goals, non-negotiable constraints are immutable and enforce critical compliance, safety, or manufacturing standards.
i.e. Maximum agitator speed in Plant A is 100 RPM, while Plant B supports 120 RPM.
Structural Causal Model (SCM)
Sales Term: Our solution to a data cold start project.
A sophisticated blueprint visually and mathematically depicts how different elements in a system influence each other. This creates a rough representation of how ingredients interact to impact outcomes, giving the ML model a pseudo-"understanding" of the product's functionality.
The SCM is automatically generated during Project Setup. It utilizes project objectives, product details (such as name and category), and user inputs and outcomes.
SCM provides an alternative route to initial data to train the ML model, allowing the Discover Solutions/Adaptive Learning process to begin without requiring tons of data or a very long initiative.
i.e. The system’s understanding of what a chocolate chip cookie entitles in the form of ingredients, processing conditions and outcomes, the relationships between these, and how modifying one input impacts others.
Metadata
The metadata is a glossary definition tied to one specific product. This includes inputs, outcomes, and other information that is used to produce recommendations. It is automatically generated by the system through the Project Setup routine.
i.e. A file which contains the different characteristics for a chocolate chip cookie, e.g. the name of the ingredients/processing conditions/outcomes, the cost, the currency of the cost, the lower and upper bounds of the input, etc. Think of it as the manual on how to read the information contained in the generated model.
Model
A graphical representation of our understanding of a product, which contains the probabilistic relationships among a set of variables. This is how the system combines information from the SCM and metadata with real-world data to build an understanding of the specific product.
i.e. A chocolate chip cookie could be represented as ingredients (flour) and dependencies (flour impacts texture).
Goals
The objectives you have set for your project. This is represented by the outcomes and the desired values you wish to obtain. There are 4 options:
- Maximize: The system will attempt to achieve the highest possible value within the defined upper range for the specified outcome.
- Minimize: The system will attempt to achieve the lowest possible value within the defined lower range for the specified outcome.
- Value: The system will attempt to set the specified outcome as close as possible to the defined target value.
- Range: The system will attempt to set the specified outcome within the defined range.
Adaptive Learning (AL)
Allows customers to leverage the custom ML Models we build them to design promising new product formulations even when the models have high predictive uncertainty. Users interact with this algorithm in Discover Solutions when they click on the Generate Recommendations and Give me More buttons.
i.e. A system-generated recommendation for a chocolate chip cookie that aims to lower the total sugar content.
Initiatives
A planned set of experiments determined by the number of formulations that can be physically tested in the lab and the target number of rounds required to achieve the desired goal.
Expected Improvement Score
The way we refer to the new UI that includes the cards view, improved AL, implemented the formulation feedback and allows users to download/upload test plans and uses that data to automatically retrains the model with the new data.
i.e. When pitching to a new customer, we can say things like these recommendations where generated by Luna.
Desirability Score
Desirability of an outcome, for a given formulation, measures how close the predicted outcome is to the outcome objective (higher is better). Total Desirability, for a given formulation, is a measure of how close the formulation outcomes are to their corresponding objectives - for all outcomes combined (higher is better). It is measured from 0 to 1.
i.e. A given set of recommendations generated by Luna AI score 0.48, 0.51 and 0.53, signaling to the user that the latter recommendation is closer to the set outcome objectives than the other two. For the chewiness level outcome, the 3 recommendations score 0.38, 0.61 and 0.45, signaling to the user that even though the third one has a better total desirability score, the second one has a much better desirability score on this particular outcome.
Formula Recommendation
A specific combination of ingredients and quantities designed to achieve the specified outcomes.
i.e. A recommended chocolate chip cookie formula that meets the temperature condition whilst maintaining liking and reducing sweetness.
Formulation Feedback
A mechanism for users to guide AI in finding recommendations based on their evaluation of generated formulations. In the UI we represent this as thumbs up and down at the ingredients and outcomes levels for each generated recommendation.
Thumbing up or down the ingredients helps the optimizer find areas to explore, or to keep exploiting current areas, whereas thumbing up or down outcomes helps the model optimize the relationships between inputs.
i.e. A user thumbs down a couple of formulations at the ingredients level, therefore the subsequent recommendations explore different values.
Optimization
The user will give a set of constraints to the model, and the model will return a single (or set) of formulations that meet that criteria, and are also maximized for the desired outcome.
i.e. I tell the model I want a cookie with at least 3oz of chocolate chips but no more than 6oz. The model will give me an arrangement of recipes that have the chocolate chips at the desired level.
Luna AI
The way we refer to the new UI that includes the project setup for low-data scenarios, cards view and timeline, improved AL, implemented the formulation feedback and allows users to download/upload test plans and uses that data to automatically retrains the model with the new data.