DISH
Turning everyday cooking into a personalised, data-driven experience – combining nutrition science, AI, and culinary creativity for healthier, more sustainable eating
DISH (Delicious, Individual, Sustainable, Healthy) reimagines home cooking through intelligent personalisation – transforming static recipes into adaptable, nutrition-aware experiences. Developed by an interdisciplinary team of technologists, nutritionists, and culinary experts, the platform enables users to tailor meals to their individual health needs, preferences, and available ingredients while safeguarding their privacy.
At its foundation, DISH brings together culinary practice and data science to create recipes that adapt to individual nutritional profiles. A machine-readable recipe language and ingredient-replacement engine allow the system to analyse, modify, and optimise dishes automatically. Each recipe is represented as a cooking graph – a structured data model that aggregates nutritional values and supports algorithmic substitutions for allergens, intolerances, or lifestyle choices such as vegan, low-cholesterol, or high-protein diets. A nutritional scoring model then evaluates how well a dish meets a user’s defined profile, computing how “healthy” a recipe is and creating a transparent bridge between scientific data and everyday cooking.
In practice, this approach enables the system to identify suitable ingredient replacements and generate tailored recipe variations – for example, adapting a classic dish to vegetarian, vegan, lactose-free, or gluten-free versions. In parallel, the project explored broader accessibility goals for users with visual, motoric, or cognitive limitations, aiming to simplify recipe steps and interfaces to make healthy cooking more inclusive. While these accessibility features remain conceptual at this stage, they set a clear direction for future development of the DISH application.
Behind this innovation lies a clear social purpose. DISH addresses three pressing challenges in modern diets: unhealthy eating habits, reduced capacity or motivation to cook, and widespread food waste. By exploring simplified recipe design, responsible ingredient replacement, and inclusive variations for different dietary and physical needs, the project illustrates how intelligent, data-driven tools can make healthier eating more accessible and appealing – validating the potential of personalised nutrition to promote sustainable everyday habits.
Built with a privacy-by-design architecture that keeps personal data on the user’s device, DISH further embodies trustworthy digital responsibility in practice. Its upcoming mobile application will extend these concepts to pilot users, demonstrating how personalised nutrition, data fairness, and culinary creativity can work together to support better food choices, reduce waste, and empower people to cook with confidence.
WHY THIS MATTERS
Healthy nutrition is a key factor in disease prevention and plays a crucial role in maintaining quality of life as we age, yet the DISH consortium recognises that nutrition knowledge alone is not enough to drive healthier eating behaviour. As they state, “nutritional information alone, as useful as it may be, is not sufficient to cook a delicious meal and meet the needs of the consumer.” While healthy nutrition is a proven factor in wellbeing, unhealthy eating habits-including reliance on ultra-processed foods-remain widespread. Many people still lack the tools or confidence to turn nutritional guidance into everyday cooking practice.
At the same time, most existing recipe and diet applications remain fragmented and commercially driven, prioritising engagement metrics over data quality or user relevance. Many treat recipes as static content rather than structured nutritional information, offering limited verification of ingredient data or flexibility for individual preferences and constraints. Applications focused on “healthy eating” typically provide fixed, pre-defined options, while others cater to niche or professional contexts, leaving everyday users without tools that adapt to their needs. In this environment, nutrition and personalisation are often treated as add-ons rather than core design principles.
In parallel, the diversity of dietary requirements and preferences is expanding rapidly. The European Dairy Association (2025) reports lactose-intolerance rates between 4 % and 56 % across EU countries; World Population Review (2023) places gluten intolerance between 0.2 % and 1.67 %; and the Good Food Institute Europe (2024) finds that nearly half of European consumers have already reduced their meat consumption, with around 40 % identifying as flexitarian, vegetarian, or vegan. These figures underline how diverse nutritional needs and preferences have become-and how important adaptable, inclusive tools are for modern cooking.
By validating a framework where recipes can be automatically analysed, modified, and contextualised for different users, DISH shows a practical pathway toward more personalised and sustainable eating. Its approach connects data fairness, culinary creativity, and user empowerment – bridging the persistent divide between nutritional science and the realities of daily life.
- European Dairy Association. (2025, January 6). Lactose intolerance in Europe [Nutrition factsheet]. European Dairy Association.
https://eda.euromilk.org/wp-content/uploads/2017/08/2025-01-06_EDA_Nutrition-Factsheet_Lactose_Final.pdf - World Population Review. (2023). Gluten intolerance by country.
https://worldpopulationreview.com/country-rankings/gluten-intolerance-by-country - Good Food Institute Europe. (2024, February 7). European consumer insights on the alternative protein sector.
https://gfieurope.org/industry/european-consumer-insights-on-the-alternative-protein-sector/
WHAT THE SOLUTION DOES
DISH transforms this gap between nutrition science and everyday cooking into a practical, intelligent system-showing how data, design, and culinary creativity can work together to make healthy eating more intuitive, adaptable, and inclusive.
Imagine Luca, a 28-year-old freelance designer who wants to eat healthier without giving up his favourite dishes. One evening, he opens the DISH app and selects a classic pasta recipe he often cooks. Instead of simply listing ingredients, the system analyses the dish in real time. It detects that the sauce is high in saturated fat and suggests balanced variations –reducing butter, adding fibre-rich vegetables, or substituting cream with a lighter alternative to meet his goal of lowering dairy and fat intake.
Behind the scenes, this process runs through the DISH Language, a machine-readable format that translates cooking instructions into data. Each recipe is expressed as a cooking graph – a structured model linking ingredients, preparation steps, and nutritional values. This enables the system to calculate precise nutrient totals, simulate substitutions, and show how each change affects flavour, preparation time, and nutritional balance.
When Luca activates his personal profile, he sets his preferences: high protein, moderate energy intake, and lower dairy consumption. The ingredient-replacement engine uses these settings to generate personalised versions of the recipe, each recalculated through the nutritional-scoring model, which evaluates how well it matches his goals. The score appears transparently, revealing how every ingredient contributes to protein, fibre, or fat levels.
For users with allergies, intolerances, or specific diets, DISH can automatically adapt familiar dishes – turning a standard risotto into a vegan, lactose-free, or lower-sodium version without losing the recipe’s logic or appeal. During prototype testing, the consortium validated this process using a set of reference recipes and dietary variants, confirming that automatic recalculation can maintain both culinary feasibility and nutritional accuracy.
All personal data and nutritional preferences remain stored locally on the user’s device, consistent with DISH’s privacy-by-design architecture. The app does not transmit individual data to external servers; only anonymised recipe information may be used to improve accuracy and quality assurance.
Through this combination of machine-readable recipes, intelligent substitution logic, and user-controlled data, DISH reimagines the recipe as a living system – one that evolves with each user’s needs, supports healthier and more sustainable choices, and demonstrates how responsible AI can make everyday cooking both personal and practical.
HOW YOU CAN USE IT
While still in its early stages, DISH provides a functional proof-of-concept that can inform developers, innovators, and researchers exploring personalised nutrition and responsible AI in food systems. Its modular design and documented methods promise to offer a basis for collaboration, reuse, and extension across multiple contexts:
For citizens and home cooks: The DISH mobile app is designed to make home cooking both smarter and simpler. Users will be able to select their favourite recipes, set personal dietary goals, and receive automatically adjusted variations that align with their preferences or health needs. Once released, this interface aims to demonstrate how adaptive recipes can support healthier choices, reduce food waste, and make nutritional balance part of daily cooking without added complexity.
For developers and digital-food innovators: The project’s key technical contributions – the DISH Language, cooking-graph logic, ingredient-replacement engine, and nutritional-scoring model-serve as a methodological reference for anyone building AI-enabled recipe, meal-planning, or food-waste-reduction tools. These components illustrate how machine-readable recipes and structured nutritional data can interact, offering a starting point for interoperable extensions or integration into existing digital-kitchen ecosystems. The consortium plans to share technical documentation and selected models once internal evaluation and licensing steps are complete.
For research and education: DISH provides a concrete case study of how responsible-by-design AI can be applied to food and nutrition. Its approach to modelling recipes as data objects, maintaining privacy-preserving local computation, and embedding nutritional transparency can inform future studies in human-computer interaction, food informatics, and applied data ethics.
Together, these outputs make DISH more than a digital recipe tool: it is an evolving methodological framework for connecting nutrition science, culinary creativity, and user agency-laying the foundation for future open, adaptive, and responsible food-tech applications.
DIGITAL RESPONSIBILITY IN PRACTICE
Privacy considerations guided the DISH architecture from the outset. The consortium adopted a local-first data model in which all user preferences, profiles, and recipe interactions are processed directly on the client device. No personal information is transmitted to central servers; only anonymised usage statistics are collected for internal testing and quality control.
User parameters-such as dietary goals, allergen flags, or nutrient limits-are stored locally within the application directory and protected by the operating system’s native encryption and security routines. For prototype evaluation, the team exported anonymised datasets that contained no user identifiers, ensuring that all test data complied with GDPR principles of data minimisation and purpose limitation. These measures demonstrate that privacy-by-design was implemented even at prototype stage, treating data protection as an intrinsic system feature rather than an afterthought.
Data fairness and reuse guide how DISH represents recipe data. The consortium defined a machine-readable DISH Language and a cooking-graph representation so that ingredients, quantities, and preparation steps are expressed as structured data that can be analysed and transformed algorithmically. From these graphs, the system computes nutritional values and supports ingredient replacement without losing the recipe’s internal logic.
To ensure consistent nutritional computation, ingredients are mapped to a reference nutrition database; for the MVP the team uses Bundeslebensmittelschlüssel (BLS) and note further alignment with the FOODON ontology and public sources for knowledge-graph modelling. Together, these choices promise to improve the consistency and traceability of nutrient data across recipes.
The team consulted the FAIR principles during requirements specification and plans to showcase data fairness in the MVP: expressing recipes in a formal grammar, parsing them into graphs, and keeping data structures and scoring logic inspectable for reuse. They also signal intent to release a universal data model and related APIs under an open-source license, supporting future interoperability and reuse within the DRG4FOOD ecosystem.
DISH ensures algorithmic trustworthiness by using deterministic, rule-based computation. Each recipe is represented as a cooking graph, in which nodes correspond to ingredients and edges describe the processing steps that connect them. Nutritional attributes are attached to each node and aggregated through the graph to compute total energy and nutrient values, creating a traceable chain between raw inputs and final composition. The nutritional-scoring model applies additive calculations to these data, producing consistent macronutrient and micronutrient totals for every variant. Because the logic is explicitly coded and reproducible, identical inputs will always yield identical nutritional results.
During development, the consortium validated the scoring functions against reference nutrition tables to confirm consistency with standard nutrient values. The algorithm also includes unit-normalisation routines to maintain coherent aggregation of ingredient data across recipe variants. By keeping all functional reasoning explicitly rule-based and so defined in code – not based on AI prediction or training – DISH makes its nutritional computation traceable, reproducible, and scientifically verifiable.
Transparency in DISH is treated as a core functional property rather than a separate reporting exercise. Each recipe variant produced by the system includes a nutritional-scoring record-a structured JSON object that lists the ingredient-level nutrient values and the aggregated totals used to generate the overall score. When the ingredient-replacement engine proposes substitutions, users can immediately see how each change affects energy, protein, carbohydrate, and fat content through direct feedback in the interface.
The consortium has also published a specification of the DISH Language and accompanying cooking-graph examples, showing exactly how recipes are parsed, transformed, and recalculated. These artefacts make the internal logic inspectable to developers and researchers, supporting independent review and reuse. By disclosing both the data structures and the transformation rules that underpin nutritional calculations, DISH ensures that transparency extends from the user interface to the underlying implementation-turning recipe computation into an open, explainable process rather than a black box.
Together these measures show how DISH applies the Digital Responsibility Goals as development principles rather than abstract ideals – embedding privacy, transparency, and fairness directly into the way recipes are modelled, processed, and shared.
CONTRIBUTION TO THE TOOLBOX
While DISH has not yet released standalone software components, its outputs provide a methodological and architectural contribution to the DRG4FOOD Toolbox. The project demonstrates how food-related data can be structured, analysed, and personalised responsibly – linking nutrition science, data design, and algorithmic transparency in a way that can inform future open-source enablers.
Key transferable assets include:
-
DISH Language Specification
A domain-specific syntax for representing recipes as machine-readable data, defining how ingredients, quantities, and preparation steps are structured and parsed within the DISH system.
Link pending [View DISH Language documentation] -
Cooking-Graph Data Model
A structured model connecting ingredients and preparation steps through nutritional attributes, enabling deterministic recalculation, substitution logic, and transparent scoring across recipe variants.
Link pending [View cooking-graph model documentation] -
Methodology and Technical Documentation
Open documentation describing the methodology for personalised recipe adaptation and nutritional scoring, including the algorithmic approach, validation routines, and privacy-by-design framework underpinning the DISH prototype.
Link pending [view methodological documentation]
The consortium intends to release these resources through the DRG4FOOD Toolbox once internal review, licensing, and quality-assurance steps are complete.
IMPACT AND OUTLOOK
The DISH project has demonstrated how responsible-by-design computation can bring nutritional science closer to everyday cooking.
Through its proof-of-concept prototype, the team is validating a complete pipeline for adaptive recipe generation – from parsing cooking instructions into structured graphs to recalculating nutritional values and substituting ingredients according to personalised dietary requirement rules. The model successfully produced vegan, vegetarian, lactose-free, egg-free, and gluten-free recipe variants during prototype testing, confirming the technical feasibility of automatic, standards-based personalisation in meal design.
Beyond its technical achievements, DISH is developing a replicable framework for data fairness, transparency, and privacy-by-design in digital food services. The project demonstrates that rule-based, locally processed algorithms can deliver traceable nutritional guidance without relying on opaque AI prediction or external data sharing.
Looking ahead, the consortium plans to publish the DISH Language grammar, parser, and scoring methodology as open documentation and scientific papers, supporting future reuse. These materials will form the foundation for follow-on work, including planned collaboration with food-technology providers, kitchen-appliance manufacturers, and nutrition platforms interested in integrating the DISH language and scoring logic into their systems.
In doing so, DISH extends the DRG4FOOD vision into the culinary domain – demonstrating how responsible AI, structured data, and user-centric design can together transform the simple act of cooking into an intelligent, transparent, and inclusive digital experience.
QUICK FACTS
- Funding
- DRG4FOOD Open Call #2
- Use case
- Digital Food Tracking
- Partners
- Start date
- Nov 2024
- End date
- Sep 2025
- Resources