Ai-powered cooking assistant with issue and improve recipe personalization methods and devices

Abstract

The invention relates to the AI-powered cooking assistant with issue and improve recipe personalization methods and devices. The AI-powered cooking assistant makes the culinary experience easier by customizing to user preferences, enabling exploring new culinary experiences, and enhancing the capabilities of kitchen appliances for optimal convenience. The AI-powered cooking assistant comprises of one cooking device (100), one smart home controller (200), one terminal device (300), and one AI engine (400). Cooking device (100) contains one connectivity module and is managed by one smart home controller via WiFi. The smart home controller (200) functions as an external server. It comprises of one database (201), one controller API (202), and one control method (203). The terminal device (300) functions as a separate device, particularly a mobile device (a phone or a tablet). The terminal device (300) comprises one user interface (301), one input method (302), one adjustment method (303), and one mutually exclusive filter (304). The AI engine (400) functions as an external server. It comprises one AI API (401), one generator method (402), one modification method (403), and one ingestion method (404).

Description

[0001] AI-POWERED COOKING ASSISTANT WITH ISSUE AND IMPROVE RECIPE PERSONALIZATION METHODS AND DEVICES

[0002] The invention relates to the Al-powered cooking assistant with issue and improve recipe personalization methods and devices. The Al-powered cooking assistant makes the culinary experience easier by customizing to user preferences, enabling exploring new culinary experiences, and enhancing the capabilities of kitchen appliances for optimal convenience.

[0003] The Al-powered cooking assistant operates in three distinct ways. It can produce personalized recipes from the input space, adapting to user preferences or the time and energy constraints they have, it can generate new, unseen recipes, allowing users to explore new tastes and cuisines and it can modify recipes to match the available ingredients and the capabilities of the cooking devices.

[0004] Cooking is a source of joy and satisfaction, yet it presents its own set of challenges. The quest for the right recipe online is often tedious and timeconsuming, as recipe websites are cluttered with advertisements and superfluous content. Moreover, the task of modifying recipes when specific ingredients are missing or when seeking appropriate substitutes is particularly cumbersome. This is a common scenario for young parents, a key demographic, who must balance multiple dishes and cater to various dietary needs, leading to food wastage due to the difficulty in adjusting recipes based on what's available in the fridge.

[0005] The primary issues are: users' desire to effectively explore new tastes and cuisines to diversify their flavor experience; the limited time and energy they have, which is further taxed by the need to navigate through multiple information sources and filter out irrelevant content and ads, a process that is both frustrating and time-consuming; and the users' preference to cook with available home ingredients and cooking devices with specific functionalities, which becomes a vexing process when they lack the required ingredients and struggle to find suitable substitutes or have to adapt the cooking parameters to available device functionalities, exacerbating the issue of food waste, especially when cooking for multiple people.

[0006] Exploring new tastes or customizing the recipes to match home ingredients or cooking device capabilities requires a sophisticated recipe personalization system. This system should be capable of generating a set of recommended or personalized recipes, which can be viewed as the target space. To create this target space, the system needs specific input information to derive conclusions. Typically, this includes the context of the user's personal preferences and their external environment. Personal preferences encompass cooking experience, current psychophysical state (e.g., mood), dietary preferences, health predispositions (e.g., allergies), and current desires (what they want to eat). The external environment comprises available tools and cooking devices, ingredients, the number of people to be served, available time, and recipe availability from various data sources. These elements constitute the input space for the recipe personalization system.

[0007] While there are existing recipe personalization systems that recommend readily available recipes or create new ones, they often have significant drawbacks. Some rely on special devices or input data that users typically do not possess or find difficult to obtain. Others focus only on specific cases within the input space, such as user health data, and fail to adapt recipes to other user's personal preferences. Additionally, adaptations of recipes to cooking device capabilities can create suboptimal solutions, as cooking operations for one specific device may not have equivalent cooking operations for another specific device. Moreover, there is a scarcity of solutions that facilitate the exploration of new culinary experiences. This limitation significantly reduces the practicality of these solutions, as they do not address the problem comprehensively and only offer partial resolutions.

[0008] The prior art involve 57 patent documents in the technical field of intelligent cooking and smart kitchen appliances, and to methods generating personalized recipes based on user preferences, recipe adjustment methods, devices, and storage mediums. The description begins by outlining patent documents that address similar challenges but employ distinct utilities or methods that diverge from our approach. These alternatives are explored to establish the context and differentiate our innovative solution. In the addition the focus narrows to three patents that bear closer resemblance to our utility configurations or methods. Here, the independent claims of these documents are scrutinized, with particular attention paid to elements that set our solution apart. This provides a detailed contrast, emphasizing the unique aspects of our system. Finaly, the prior art documents offers an examination of existing solutions, highlighting the key differences that underscore the advantages and novelty of our proposal.

[0009] Five irrelevant documents are inactive and thus not pertinent to the current state of the art (CN111863192A, CN111863193A, W02019223300A1 , WO2020181722A1 , WO2018023272A1 ). The document US10387499B2 doesn’t relate to the technical field of recipe generation, as it describes the supply method. Document WO2021110066A1 discusses a food maturity identification method. EP4279847A1 is focused on improving the quality of the image of an object in a refrigerator. Documents W02023105033A1 and US11606619B2 relate to the technical field of display unit and do not relate to the technical field of the recipe system.

[0010] A significant number of documents describe systems and methods reliant on a refrigerator or a cooling device (EP4279847A1 , EP2985553B1 , US11521391 B2,

[0011] US9784497B2, CN110875090A, CN 116434912A, US10969162B2,

[0012] CN1 15080832A, CN105509394B, CN1 17520640A, CN107610750A,

[0013] CN1 15080831 A, CN115371326B, CN1 16051223A, CN113220977B,

[0014] CN1 13299367A, US11732961 B2, CN107863138B, CN 113569140A,

[0015] CN1 11105862A, US11531910B2), which our solution does not require.

[0016] Document WO2022 / 231174A1 describes an electronic device with a camera to photograph the tray, a feature our solution does not incorporate. CN115581398A describes a mechanical arm, which is not included in our solution. Document US20240041251A1 outlines a specific cooking device comprising curry vessel, ingredient rack, a rice vessel and two induction cooktops, which is not analogous to our Al-powered cooking assistant.

[0017] Document CN114864047A describes a recipe recommendation method based on recognizing the specific user by audio to create a more accurate recommendation. Our system does not employ user recognition from an audio source; instead, we use audio as an alternative input mechanism 306. US11322149B2 details a method to automatically extract a recipe using an image or speech from a cooking video. This is different from our approach, which does not rely on such extraction methods. Document US20240144173A1 presents an online concierge system that suggests recipes to reduce grocery item waste. Unlike this system, our assistant does not just identify the best available recipes; it can generate new ones. CN117936027A discloses a recipe generation method that includes getting the user’s facial complexion feature to determine the physiological condition. Our solution does not incorporate such methods; instead, our input method 302 considers manual input by the user. CN110287306B uses a food knowledge graph to recommend recipes even when certain ingredients are lacking. Our solution does not contain a knowledge graph methodology. CN115248566A describes a method of different recipe albums corresponding to different modes of the cooking appliance. CN116643502A addresses the impact of ingredient quantity and cooking parameters on the final taste, colour, and texture by obtaining user feedback and determining at least one parameter to be adjusted according to its weight. Our method does notinclude considerations of weight influencing the final characteristics of the food.

[0018] US2022183330A1 provides a solution for users intending to cook several recipes simultaneously. It guides users on an optimized cooking order and automatically controls the cooking apparatuses used. However, our Al-powered cooking assistant does not provide such a guide or control cooking apparatuses.

[0019] WO2024068767A1 publication addresses the synchronization of recipes between different devices, even across different brands. It claims a method of obtaining the progress and state of the recipe when the user switches between devices. Our Al-powered cooking assistant does not involve such synchronization between devices.

[0020] CN110289078A discloses a method for improving recipe recommendations by considering the actual cooking ability of the user, determined from video images of cooking behavior. While similar to our adjustment method 303, our system does not determine cooking ability from video images but rather relies on manual input from the user. Document US11610665B2 describes a method for improving food-related personalization where food-related preferences are represented by a vector in the recipe vector space. Constraints in this space then define a set of recipes appropriate for the user.

[0021] Similarly, US11587140B2 determines ingredients vectors as the basis for determining substitutes and creating a personalized food plan. Both methods solve a similar problem with a machine learning technique and are associated with our input method 302 combined with the generator method 402. However, their approach is different as they use a trained neural network to map food- related preferences and ingredients to a vector space and then derive conclusions based on similarity. This is a discriminative approach. In contrast, our solution does not map input data to a vector space but uses the knowledge of a model to generate the result, making it purely a generative approach.

[0022] US10162481 B2 also describes a method like our input method 302 combined with generator method 402, where various user information is received via an interface to create an end food or drink product with desired characteristics. However, their method involves ranking ingredients using gas chromatography and mass spectrometer data, which is not a part of our approach.

[0023] CN116564472A aims to improve recipe recommendation by considering the state of the equipment the user owns. This aligns with our control method 203. However, their method is used to obtain a recipe list associated with the equipment status, whereas our ingestion method 404 modifies the recipe itself based on this information.

[0024] Document CN112369122B describes operating a cooking device based on adjusted cooking parameters extracted from the obtained recipe and adjusted by the user. For instance, if the cooking device is a steam oven, the user is not limited to only recipes for steam oven, but may use, for example, a microwave oven recipe, since the cooking parameters are converted to more suitable alternatives. It also records the cooking process's operating data for gradual improvements. Our ingestion method 404 adjusts the recipe to be associated with the cooking device 100 in a rather different way. Their method converts cooking parameters from one device-specific cooking operations to other device-specific cooking operations. In contrast, our method uses device-agnostic to devicespecific mapping. Moreover, it allows the user to select a cooking operation 101 among the capabilities of the selected cooking device and convert the whole recipe, not just the cooking parameters, to be suitable for the selection. Furthermore, the ingestion method 404 does not include recording operating data for improvements.

[0025] Document EP3366175A1 offers a customized diet prepared by a cloud computer server based on physiological parameters acquired by measuring devices. Our system, however, does not utilize measuring devices nor measure physiological parameters such as glucose levels or heartbeat.

[0026] Similarly, CN117747061 A describes a menu recommendation method that requires acquiring body parameters to characterize obesity levels.

[0027] CN106021564B outlines a method for sending a list of recommended recipes formed according to the user's eating habits information. While it may resemble our input method 302 in obtaining user information, this method does not generate recipes but sorts them according to dietary habits and outputs the recipe with the highest number of ingredients matching those habits.

[0028] CN110942815A assesses the health status of the body and recommends recipes, accordingly, necessitating a medical examination device. CN117174251A provides a health management data method that obtains the user's physical parameters, analyzes them to obtain risk parameters, and generates health management data based on these risk parameters. Our method does not involve such analysis or data generation.

[0029] Document CN114512214A describes a cooking method and device related to health. According to their first claim, they obtain user health data and generate a healthy diet recipe. The user then selects the target recipe on the terminal device. Based on the selection, a cooking instruction is generated and sent to a cooking device. This approach is also used by our method, namely, input method 302 combined with the generator method 402 and control method 203. However, their method is distinct from ours in two ways. First, they define the health data as current physical health state that can include blood glucose meter, a body fat scale, and a blood pressure meter, among others. We don’t include any such data. Second, their claim 10 specifies that the user's current health condition is acquired by a health detection device. In contrast, our solution relies solely on health conditions manually input by the user, thus eliminating the need for a secondary, specialized device that users typically do not possess.

[0030] Document CN112817237A contains four independent claims: 1 , 10, 14, and 15 with the initial claim disclosing a cooking control method that finds a list of recommended recipes based on the provided ingredients. It then selects a target recipe and cooking device, determining cooking parameters based on the device chosen and the quantity of ingredients in the recipe. This process bears resemblance to our methods, which include the input method 302, the generator method 402, the ingestion method 404, and the control method 203. However, our methodology diverges in several key aspects, (i) Our input method is more versatile, capturing not only ingredient information but also additional input space features such as dietary preferences, health needs, and cuisine style, (ii) The patent claims the cooking control device determines recommended recipes, but step 102 indicates it merely searches for recipes online. In contrast, our smart home controller 200, akin to their cooking control device, does not determine recommended recipes. This function is a feature of our Al engine 400, which not only searches a set of basic recipes 501 to consolidate knowledge and enhance results but also generates a new set of recipes from its knowledge base. (iii).

[0031] Unlike the method described in the patent, our methodology does not send the target recipe to the cooking device; it only sends the cooking parameters.

[0032] Claim 10 describes the cooking control method where the cooking control device determines the target recipe and the reference cooking parameter. It also includes a detection feature for a second food material within the cooking area, which, if corresponding to the target recipe, prompts the device to cook the ingredient. Our method diverges as it does not employ detection to identify food materials in the cooking area.

[0033] For claim 14, the described cooking control device significantly differs from our smart home controller 200. It features four distinct units responsible for determining the first food material, a recommended recipe, a target recipe, a target cooking device, and reference cooking parameters. In our system, the user determines the food material, target recipe, and cooking device, while the recommended recipe and reference cooking parameters are determined by the Al engine 400. Additionally, unlike their device which sends both the target recipe and reference cooking parameters to the cooking device, our smart home controller 200 only transmits the reference cooking parameters.

[0034] Claim 15 details a target cooking device equipped with a receiving unit, an identification unit, and a processing unit. Our corresponding device does not include an identification unit to detect and identify a second reference food material placed in the cooking area.

[0035] CN114048375A presents a recipe recommendation method that gathers information on available ingredients, user dietary preferences, and health status to generate a recommended recipe from the intersection of recipe sets derived from different data sources. It includes two independent claims 1 , and 10.

[0036] This method shares similarities with our input method 302 and the mutually exclusive filter 304. However, our approach is distinct in two key ways, (i) Our method searches for the solution within the input space (i.e., input information), as opposed to their method that seeks intersections within the target space, or recipes. While it might be argued that our generator method 402 also utilizes an intersection of recipes when searching for basic recipes 501 for consolidation context, the differences are notable. Firstly, we use the entire set of matched basic recipes 501, not just their intersection. Secondly, each basic recipe 501 is assigned an additional weight factor. Thirdly, the basic recipes 501 that form the consolidation context are integrated into the input space and do not form part of the final target space, which is the personalized recipes 502. (ii) Our method uses human-computer interaction approach that empowers the user by explicitly displaying the mutual exclusiveness within the input space. Conversely, their method implicitly eliminates incompatible recipes, thus removing them from the user's consideration without explicit indication.

[0037] US11631010B1 details a system of connected kitchen appliances and a dynamic recipe generation method. This method can convert user-friendly recipes, typically found in physical cookbooks, into a data structure known as an abstract recipe. This abstract recipe can then be adapted into various types and formats, serving either the cooking devices as a functional recipe or the user as a tangible recipe. The document asserts three independent claims: 1 , 19, and 20.

[0038] Claim 1 could be associated with our terminal device 300, Al engine 400, the smart home controller 200, and the cooking device 100, yet there are notable distinctions, (i) Their system includes actions referring to operations by the user or cooking device, stored in an action database, associated with kitchen appliances in an appliance database, and integrated into recipe steps as action associations by a transformation engine. Our system does not incorporate a comparable feature related to actions, (ii) Their system features an ingestion engine that converts user-friendly recipes into abstract recipes, which our system does not require as our database 201 already contains recipes in the appropriate format, (iii) Their transformation engine scales ingredient quantities and identifies kitchen appliances for recipe steps, akin to our ingestion method 404. However, our method does not scale ingredient quantities and adapts the recipe to the cooking device rather than just identifying appliances, (iv) They have a tangible recipe generation engine that converts functional recipes into user-friendly formats for the interface, a method our system does not employ. Like claim 1 , claims 19 and 20 of US11631010B 1 also present features that are not present in our system. Claim 20 describes a client device that includes a user interface, a transformation engine, and a tangible recipe generation engine. Our terminal device 300 does possess a user interface; however, it does not incorporate a transformation engine or a tangible recipe generation engine.

[0039] Our method differentiates itself from two distinct groups of competitors in the cooking and smart kitchen space. On one side, there are Al-powered recipe apps like Food Mood, SideChef, ChefGPT, and Yummly that provide recipe recommendations and meal planning capabilities. They can be seen as like our Al engine 400. However, our Al engine 400 contains a distinct feature of providing recipes based on the specially designed cooking profile 505, and consolidation context, utilizing prior knowledge to provide better results. Moreover, these apps do not offer seamless integration with kitchen appliances or cooking devices.

[0040] On the other hand, white appliance manufacturers like LG, Samsung, and GE offer smart home platforms and connectivity solutions for their appliances but do not match the advanced Al-driven personalization and recipe generation that our method offers. Even if they provide some Al-driven recipe generation, our method includes specific components that make it superior, such as mutually exclusive filter, a particular cooking profile structure, consolidation context and a modification method with swapping and quick-fix functionality.

[0041] The detailed description of our solution begins with utility where the utility is elaborated through an in-depth explanation of the mechanical configuration of the system, including the cooking device, smart home controller, terminal device and Al engine. Following this, in adition deilve into describing recipe forms and acquiring cooking profiles, detailing the input and adjustment method to capture the content for the Al engine. It also continues with describing how recipes are created and modified. Finally, we provides a comprehensive description of interaction between the components of the whole system. This comprehensive breakdown ensures a clear understanding of the innovative approach our solution offers in the culinary technology space.

[0042] The present application provides a method and system which provides a dynamic approach to recipe generation that is customizable to users and which allows the integration of connected appliances into the cooking process, with issue and improve recipe personalization methods and devices. The Al-powered cooking assistant makes the culinary experience easier by customizing to user preferences, enabling exploring new culinary experiences, and enhancing the capabilities of kitchen appliances for optimal convenience.

[0043] The Al-powered cooking assistant operates in three distinct ways. It can produce personalized recipes from the input space, adapting to user preferences or the time and energy constraints they have, it can generate new, unseen recipes, allowing users to explore new tastes and cuisines and it can modify recipes to match the available ingredients and the capabilities of the cooking devices.

[0044] The present application will now be described in association with the accompanying drawings in which:

[0045] FIG. 1 is an diagram of the Al-powered cooking assistant;

[0046] FIG. 2 is an smart home controller 200;

[0047] FIG. 3 is an terminal device 300;

[0048] FIG. 4 is an Al engine 400;

[0049] FIG. 5 is an forms of the recipe;

[0050] FIG. 6 is acquiring the cooking profile 505;

[0051] FIG. 7 is the input method 302;

[0052] FIG. 8 is the adjustment method 303;

[0053] FIG. 9 is the generator method 402;

[0054] FIG.10 is the modification method 403;

[0055] FIG.11 is the ingestion method 404;

[0056] FIG.12 is an interaction when creating personalized recipes 502;

[0057] FIG.13 is an interactions when crating modified recipes 503;

[0058] FIG.14 is an interactions when creating functional recipes 504; FIG.15 is an Interactions when controlling the cooking steps by the smart home controller 200;

[0059] The Al-powered cooking assistant is a system of at least one device with a processor and a memory. It comprises at least one cooking device 100, at least one smart home controller 200, at least one terminal device 300, and at least one Al engine 400, as depicted in FIG.1 . The cooking device 100, such as an oven, is configured to cook the dish by applying heat and contains one or more connectivity modules that enable the device to be managed by the smart home cotrollers 200 via network (e.g., WiFi).

[0060] The smart home controller 200 is a comprehensive device that is configured to manage smart home appliances. It comprises at least one database 201, at least one controller application programming interface (API) 202, and at least one control method 203, as shown in FIG. 2. The smart home controller 200 may be integrated as a controller within the cooking device 100 or function as an external server.

[0061] Database 201 is configured to store information for at least one ingredient, cuisine, taste, psychophysical state, dietary preference, allergy, cooking profile 505, inventory, and basic recipe 501.

[0062] The controller API 202 allows a common interface for communication with the cooking device 100 or the terminal device 300.

[0063] Control method 203 is configured to control the appliances, update them, or retrieve information about their status, among others. What is more, it communicates with the terminal device 300 by getting request commands and sending information back to the terminal device 300. Control method 203 also offers advanced control capabilities by controlling the cooking steps of the functional recipe 504. In this regard, it can reduce cooking time and ensure optimal cooking conditions. For example, the functional recipe 504 contains a cooking step of preheating the cooking device to 200°C. The control method 203 can start preheating the cooking device 100, so that when that cooking step is reached in the recipe, the cooking device 100 is already hot. At the same time, it can send preparations steps to the user via the terminal device 300.

[0064] For example, it can send a step of slicing the chicken breast into small strips.

[0065] Database 201 is configured to store information for at least one ingredient, cuisine, taste, psychophysical state, dietary preference, allergy, cooking profile 505, inventory, and basic recipe 501.

[0066] The terminal device 300, as shown in FIG. 3, can be any device with a display that enables the user of the smart home system to manage it. The terminal device 300 can be integrated within the cooking device 100 or function as a separate device. For example, it can be a mobile phone, a tablet, a computer, a TV, among others. The terminal device 300 comprises at least one user interface 301 , at least one input method 302, at least one adjustment method 303, and at least one mutually exclusive filter 304.

[0067] The user interface 301 provides a graphical user interface allowing a user to interact with the system. The input method 302 and adjustment method 303 provide the means to capture cooking profile 505 for generating personalized recipes 502. The mutually exclusive filter 304 is used to correct the discrepancies of the content provided by the input method 302 and the adjustment method 303 and maintain the integrity of personalized recipe generation and avoid contradictory recipe suggestions. The input method 302, adjustment method 303, and the mutually exclusive filter 304 are described in more details in the Acquiring Cooking Profile. The terminal device 300 is configured to communicate with the smart home controller 200 and with the Al engine 400. When communicating with the smart home controller 200 it can send recipes to save them for later use, request list of available devices, or send provisions from the user to prepare the dish according to the functional recipe 504. When communicating with the Al engine 400, the terminal device 300 can send provisions that are needed for successful recipe creation or request personalized recipes 502, modified recipes 503, or functional recipes 504.

[0068] The Al engine 400, as shown in FIG. 4, comprises at least one Al API 401, at least one generator method 402, at least one of modification method 403, and at least one ingestion method 404.

[0069] The Al API 401 allows a common interface for communication with any type of terminal device 300. The generator method 402, modification method 403, and ingestion method 404 are further described in Acquiring Cooking Profile.

[0070] The Al engine 400 can be integrated within the cooking device 100 or function as a separate device, for example, an external server. The Al engine 400 is configured to communicate with the terminal device 300 by requesting provisions needed for recipe generation and sending different types of generated recipes back to the terminal device 300. The Al engine 400 is also configured to send the functional recipe 504 to the smart home controller 200 to control the cooking steps and cook the food to the final dish.

[0071] There are four types of recipes form: basic recipe 501 , personalized recipe 502, modified recipe 503, and functional recipe 504.

[0072] The basic recipe 501 is a set of instructions for preparing a particular dish, including a list of the ingredients required. It is characterized in that it is stored in database 201 and isn’t necessarily associated with the cooking profile 505 nor the cooking device 100. In other words, the basic recipe 501 contains ingredients and a sequence of cooking steps, however, the ingredients may not be the ones the user prefer. For example, the recipe may contain chicken, but the user is vegan. Similarly, the sequence of cooking steps may not contain steps, cooking parameters, or cooking operations 101 that are specific to the cooking device 100. For example, there aren’t any steps of baking with steam that would be possible with a steam oven. The basic recipe 501 is, thus, user and device agnostic. The device agnosticism is accomplished by including only cooking steps, cooking parameters, or cooking operations 101 that are generic and are available to all cooking devices configured to cook by applying heat. The basic recipe 501 can be obtained from various data sources. This includes but is not limited to online web data sources (e.g., websites with different recipes, recipe databases and datasets), user data sources (e.g., cooking books, recipes that the user knows by heart), recipes from a cooking device or the ones available on any other device, or sources provided by the Al engine 400 such as personalized recipes 502 or modified recipes 503.

[0073] The personalized recipe 502 is generated by the generator method 402 and is associated with the cooking profile 505. However, it is not associated with the cooking device 100. For example, the personalized recipe 502 contains spicy ingredients, since the user prefers spicy taste. However, it may not contain the cooking steps with capabilities that can be performed with the selected cooking device 100. The personalized recipe is user-specific and device-agnostic.

[0074] The modified recipe 503 is like the personalized recipe 502 in a sense that it is user-specific and device-agnostic (i.e., it is associated with the cooking profile 505 and is not associated with the cooking device 100). However, it is generated by the modification method 403 and is associated to the modified cooking profile 505 or the modified ingredients of the recipe, whether it is the basic recipe 501 or the personalized recipe 502. For example, the user prefers a high protein food and the personalized recipe 502 contains 100 g of mozzarella. However, the user may not have mozzarella at home and would like to modify the personalized recipe 502 by substituting mozzarella with gouda, which he has at home. The modified recipe 503 would then contain all the ingredients and actions from the personalized recipe 502, except that 100 g mozzarella would be replaced by, for example, 150 g gouda.

[0075] The functional recipe 504 is generated by the ingestion method 404 and is associated with the cooking profile 505 and the selected capabilities of the selected cooking device 100. That is, the functional recipe 504 contains all the ingredients the user prefers including a sequence of cooking steps, cooking parameters, or cooking operations 101 that are adapted to the selected cooking operations 101 of the selected cooking device 100. For example, the recipe that will be transformed by the ingestion method 404 contains an action of generic baking on 200°C. Since the user has a cooking device with a ventilation function, the ingestion method 404 may transform the action to baking with the ventilation system on 180°C if the user chooses the particular capability of the cooking device.

[0076] The forms of recipe will now be described in the context of producing the recipe for use by a user with reference to the Al engine 400, as depicted in FIG. 5. The generator method 402 generates at least one personalized recipe 502. The user can explore various personalized recipes 502 and respond in three different ways, (i) The user can discard the personalized recipe 502. (ii) The user can save the personalized recipe 502 in database 201. (iii) The user can select the personalized recipe 502 for further processing. Further processing can be modifying the personalized recipe 502 by the modification method 403 or transforming the personalized recipe 502 by the ingestion method 404 to get the functional recipe 504.

[0077] In the context of the forms of the recipe, the basic recipe 501, can be created by saving the personalized recipes 502 or the modified recipes 503 into database 201. However, as previously described the basic recipes 502 can also be obtained from other data sources. The basic recipe 501 can be transformed into the modified recipe 503 or the functional recipe 504.

[0078] Considering the modified recipe 503, it can be generated by the modification method 403 from the personalized recipe 502 or from the basic recipe 501. The modified recipe 503 can be further processed by the ingestion method 404 to create the functional recipe 504 or it can be saved to database 201.

[0079] According to the different functionalities of different types of recipes, the types of recipes can be categorized according to the input space and target space. Depending on the aim of the user, the input space can be made up by all the recipe types except the functional recipe 504. On the other hand, the target space can include all the recipe types except the basic recipe 501.

[0080] The primary issue when cooking is to get the personalized recipe 502 that is customized in the context of the user’s personal preferences and his external environment, i.e., the input space. Furthermore, there are various mechanisms for how to acquire the input space. The context may be provided in the form of text, conversations, selections, visual and audio input, and others. These types of input mechanisms can create content in various formats that can be incompatible between each other or in relation to the methods ingesting them for further processing. Therefore, to get the right recipe it is necessary to first create a structure that will be in line with the input space and will be comprehensible by methods no matter the content’s form. The cooking profile 505 is a combination of personal preferences and constraints from the external environment characterized in that it has a specific structure or format compatible by the required format used in all of the Al engine 400 methods. For example, the format can be a structured JSON format, a LLM prompt, a data structure, a table in the database, among others. The cooking profile 505 may contain preferences of cuisines, taste, dietary needs, allergies, disliked ingredients, among others. The content for the cooking profile 505 is provided by the input methods 302 and the adjustment methods 303. However, not all content combinations are appropriate or compatible. For instance, if the dietary preference suggest vegan, but the meat was selected as an ingredient input content 305, these two choices are incompatible or mutually exclusive. The mutually exclusive filter 304 checks for such discrepancies between the content provided by the input methods 302 and the adjustment methods 303 and corrects them with human-computer interaction. It explicitly indicates the discrepancies to the user to get the user’s consideration into account before the final correction. In contrast to implicit elimination of discrepancies, this method better reflects the user's true preferences and restrictions. The cooking profile 505 is then the final output of the mutually exclusive filter 304, as shown in FIG.6. The disclosed feature of the mutually exclusive filter 304 is crucial for maintaining the integrity of personalized recipe generation and avoiding contradictory recipe suggestions.

[0081] To create the content for the cooking profile 505, several types of input methods 302 can be used. The input methods 302 are divided into input content 305 and input mechanism 306, as depicted in FIG.7. Input content 305 refers to the actual information that is being input, what the input method captures (e.g., dietary preferences, ingredients). The input mechanism 306 refers to the way this content is captured, such as through typing, voice recognition, or touch. The input method can include one or more contents with any combination of one or more mechanisms.

[0082] The input method 302 can capture several different input contents, (i) The user can start generating personalized recipes 502 by using a random generator. The random generator randomly selects ingredients. The random generator can be configured as a slot machine with at least 1 reel slot. If multiple reel slots are present, they can also be locked if the ingredient satisfies the user needs. In another variant, reel slots are configured to show specific categories of the ingredients. For example, one of the reel slots could show only protein ingredients and the other only vegetables, (ii) Another input content 305 is psychophysical state. It refers to the interrelation of the psychological and physical aspects of the user’s state of being, including mood and emotional state. The psychophysical state influences the user’s preferences towards certain personalized recipes 502. For example, the user could input that he feels happy, tired, or hungry and that would affect the generation of the personalized recipe 502. (iii) Another input content 305 is dietary preferences. Dietary preferences refer to innate predispositions to favor or avoid certain food groups. It includes but is not limited to different diets (e.g., low-carb, gluten-free, vegan, mediterranean). For example, the user can enter meat or vegan dietary preferences, (iv) Another input content 305 is health needs. They refer to innate predispositions to favor or avoid certain food groups because of their health status (e.g., allergies). For example, the user can enter allergies, where certain ingredients must be avoided, (v) Another input content 305 is taste. It refers to sensation and perception of flavors in food and an individual’s preference or inclination towards certain flavors. For example, the user can determine a more sour, sweet, or salty taste, (vi) Another input content 305 is cuisine. The cuisine is the style of cooking characterized by distinctive ingredients, techniques, and dishes, usually associated with a specific culture or geographic region. For example, French cuisine is known for its use of herbs and wine, while Chinese cuisine is famous for its diverse flavors and use of soy sauce and rice vinegar. The user can in this regard choose among multiple cuisine styles (e.g., Italian, Chinese, or Mexican cuisine style), (vii) Another input content 305 is ingredients. Ingredients are any of the food or substances that are combined to make a particular dish. They can range from fresh produce and spices to meats, grains, and dairy products. The user can input one or more ingredients as a starting point for generating personalized recipes 502. Since not all recipes can contain all the selected ingredients, the ingredients are optional by default. This means that the Al engine 400 has the freedom to omit some of the ingredients in the personalized recipes 502. To avoid this, the user can specifically lock ingredients that he / she wants to be included as a necessity, (viii) Another input content 305 is cooking profiles 505. Since the cooking profile 505 already describes the context for creating personalized recipes 502, it can be used as a part of a predefined input content 305. The benefit of this approach is that the user doesn’t have to add his preferred input content 305 every time he uses the Al-powered cooking assistant. For example, the user could create a weekend profile that would be configured to his / her preferences for weekends. Another example would be a party profile that may include preferences normally considered for a party. The cooking profile could also refer to a person. For example, a profile of a friend that the user often cooks and has certain allergy restrictions. The user can provide one or more cooking profiles 505. (ix) Another input content 305 is inventory. It refers to the stock of ingredients available in stock at home, in cupboard, countertop, pantry, drawer, among other things. For example, spices, oils, sauces, canned and jarred foods. The inventory can be prefilled with ingredients that are commonly found in the country of the user, (x) Another input content 305 is basic recipes 501, already described in Section 3.2. Similarly, to cooking profiles 505, the basic recipes 501, can be used to save user time. The user can provide one or more basic recipes 501.

[0083] Any content (input content 305 or adjustment content 307) can be captured by several different input mechanisms 306. (i) Content can be captured via text. The user can write the content with combination of autofill, (ii) Another input mechanism (6) is selection. The user can use a provided list of contents, where he can search and select one or more contents. For example, if the content is inventory, the user can select ingredients that he has at home, (iii) Another input mechanism 306 is visual input. The user can take a picture of the content. The content can be any previously described input content 305. He can take one picture of all contents at once, or he can take multiple pictures with one or more available contents. For example, the user can take a picture of ingredients on the table. In another example the user can take a picture of a bill that contains bought groceries. In another example the user can take a picture of a recipe from a book. The user can also use visual gestures such as facial movements, hand gestures or body gestures to provide the content, (iv) The user can also use audio input as input mechanism 306. He can use speech and name one or more contents, or he can also describe them. If description is used the method will provide the named content suggestions that can be selected. The user can also use audio signals (e.g., music) to provide emotional cues as content, (v) Another input mechanism 306 is based on haptic. It refers to interaction through touch by applying force and pressure or using touch gestures.

[0084] There are also several types of adjustment methods 303 that are divided into adjustment content and input mechanism, as depicted in FIG 8. The adjustment content 307 refers to the actual information that is being input, what the adjustment method 303 captures. The input mechanism 306 refers to the same mechanism that is used for the input method 302. The adjustment method 303 can include one or more adjustment contents 307 with any combination of one or more input mechanisms 306. The type of input mechanism is not dependent on the type of input mechanism that is used for the input method.

[0085] Three basic and arbitrary additional adjustment contents 307 can be provided.

[0086] (i) The user can provide time as an adjustment content. The time indicates how much time is user able to spend on cooking the meal, (ii) Another adjustment content 307 is effort. It refers to the user’s available energy he can put into the cooking and recipe’s difficulty. How much preparation is needed, how complex are the cooking steps and tools that are needed, how much cleaning is needed after the cooking is finished. The visual depiction of the effort can vary, but the optimal method is to present it as a category of three levels: low, medium, high. For low effort, recipes require minimal preparation and use basic cooking techniques. Focus is on simple, easy-to-follow instructions and common ingredients. For medium effort, recipes require moderate preparation and use intermediate cooking techniques. It includes a reasonable number of ingredients and steps that build on foundational cooking skills. For high effort, recipes use a variety of ingredients and provide detailed instructions for preparing more complex dishes, (iii) Another adjustment content 307 is skill. It refers to the user’s cooking knowledge and cooking experience, (iv) Arbitrary additional adjustment contents 307 can also be provided. For example, type of dish (cold or warm), seasonal (yes, no), price (low, medium, and high), among others.

[0087] The Al Engine 400 comprises four main methods the generator method 402, the modification method 403, and the ingestion method 404. There are also four complementary methods the input method 302, adjustment method 303, mutually exclusive filter 304 and control method 203. The complementary methods are part of the terminal device 300, except for the control method 203, which is part of the smart home controller 200. The complementary methods were already described in previous sections.

[0088] There can be at least one of each of the methods, but normally there is only one of each. However, the user can use multiple input methods 300 at the same time

[0089] As depicted in FIG. 9, the generator method 402 is configured to generate the suggested recipes 502 from cooking profiles 505 that were acquired from the user. What is more, the generator method 402 can also use basic recipes 501 to provide better personalized recipes 502. This is done by extracting keywords and semantics from the cooking profiles 505 and searching for associated recipes among provided basic recipes 501. The results provide a consolidation context which acts as a prior knowledge how the relation and composition of ingredients and their characteristics affects the characteristics of the final dish.

[0090] The generator method 404 comprises at least one rephrasing method 405, at least one keyword extraction method 406, at least one keyword search method 407, at least one question embedding 408, at least one semantic search 409, at least one context consolidation method 410, at least one augmentation method 411, and at least one model 412. The rephrasing method 405 transforms the cooking profile 505, if necessary, so that intent and objective of the user’s context becomes clearer. For example, if the user uses an audio input mechanism 306, the cooking profile 505 may contain conversational content that is not associated with the recipes. Rephrasing method 405 may extract the appropriate information specific for the recipe creation. The output of the rephrasing method 405 is used as an initial objective and as an embedding for the semantic search 409 and as input for the keyword extraction method 406.

[0091] The keyword extraction method 406 extracts keywords from the rephrasing of the cooking profiles 505. For example, it may extract all the ingredients. The extracted keywords are then used in the keyword search method 407 that searches for the basic recipes 501 containing the provided keywords.

[0092] The semantic search 409 uses initial objectives from the rephrased cooking profiles 505 and embeddings from the question embeddings 408 to search for basic recipes 501 containing similar context. The semantic search 409 is an addition to the simpler keyword search method 407 that may fail to find appropriate basic recipes 501. For example, the user may input pig as an input content. The keyword search method 407 may fail to find appropriate basic recipes 501 since the keyword pig is normally not found in the recipes. However, the user probably wanted a recipe with pork. Since the semantic search 409 searches the basic recipes 501 according to the semantical meaning of the animal pig it may find basic recipes 501 that associate to the input content of the user.

[0093] Since the keyword search methods 407 and the semantic searches 409 can result in different set of basic recipes 501, the context consolidation method 410 is used to create a consolidation context. The consolidation context is a set of all basic recipes 501 found by either of the search methods with an additional weight factor that corresponds to how good was the match between the search and the basic recipe 501. The weight factor can be given by a probabilistic inference of a Bayesian network

[0094] P(R = r(:K = kfS = s, C = c) = p(c) p(s [ c) p(k c) p(r [ s) where is a random variable representing the basic recipe 501, A" is a random variable representing keyword search method 407, 5 is a random variable representing the semantic search 409, and C is a random variable representing the rephrased cooking profile 505.

[0095] The consolidation context is augmented with the initial cooking profile 505 and the rephrased cooking profile 505 by augmentation method 411 and used as an input to the model 412 that generates the personalized recipes 502. The model 412 can be any generative method, including but not limited to diffusion models, generative adversarial networks, autoencoders, transformers, neural radiance fields (NeRFs). However, the best approach is to use large language models including but not limited to autoregressive language models, transformer-based models, encoder-decoder based models, encoder only based models, decoder only based models, and multimodal or hybrid models that combine different modalities. Modalities correspond to described input mechanisms 306.

[0096] The user can modify the recipes by using the modification method 403, presented in FIG 10. For example, the user can adjust the quantity of ingredients, adding, or removing ingredients, or changing the cooking time and effort levels. The modification method then adjusts the recipes accordingly.

[0097] The modification method 403 is configured similarly to the generator method 402 since it uses similar input components and the same algorithms. However, some components can be missing, since the modification method 403 only considers the modified content. For example, in the modified cooking profile 505 there may be missing allergy information, since this was already communicated by the user previously.

[0098] Additionally, the modification method 403 has a swapping functionality that allows user to substitute ingredients with similar ones. When using the swap function, the user gets a list of substitutes. Each substitute can contain additional information that can help decide if the substitute is appropriate.

[0099] Additionally, the modification method 403 has a quick fix functionality that allows the user to modify recipes with just one press. The recipe can be changed into one of the fix-options. For example, fix options may be high protein modifies the personalized recipe to contain at least 20 grams of protein; vegan modifies to a plant-based recipe; simplify reduces ingredients and the steps; gluten free excludes foods that contain glute. There can be an arbitrary number of fixoptions. The best embodiment contains four fix-options.

[0100] When the user selects any kind of recipes for cooking, whether they are the basic recipe 501, personalized recipe 502, or modified recipe 503, the recipes are device-agnostic (i.e., they are not associated with the capabilities of the cooking devices 100, where the dishes will be cooked). However the selected recipe can be cooked in different ways, by different cooking operations 101 of the selected cooking device 100.

[0101] There exist solutions that adapt recipes to cooking devices but are limited to obtain a list of recipes associated with the equipment or use complex adaptation methods that convert cooking parameters from one device-specific cooking operations to other device-specific cooking operations. These complex adaptations can create deficient recipes, since the adaptation is not a one-to-one mapping. In other words, cooking operations for one specific device may not have equivalent cooking operations for another specific device. For instance, a microwave cooking operation has no equivalent in the convectional oven.

[0102] To overcome this, the ingestion method 404 that incorporates the humancomputer interaction methodology is used to adjust the recipes to specific capabilities of the cooking devices 100, as depicted in FIG. 11. It is specifically designed to adapt the recipe with the method of mapping from device-agnostic cooking operations 101 to device-specific cooking operations 101.

[0103] The user must first select at least one cooking device 100 from the list of available devices that was provided by the smart home controllers 200. The user then continues by selecting particular cooking operations 101 available on the cooking device 100. The characteristics of the cooking devices 100 are then send alongside the selected recipes and selected cooking operations 101 to the ingestion method 404 that adjusts the recipes, as depicted for instance, the user normally cooks vegetables with convection cooking operation 101. He decides, for example, to cook the same recipe with steam cooking operation 101, since he has a steam oven that has such functionality. The user follows the ingestion method 404 and provides the appropriate input information so that the ingestion method 404 modifies the recipe accordingly.

[0104] The ingestion method 404 is configured to adjust the recipe as a whole, not just the cooking parameters of the cooking device 100. In this regard, the ingestion method 404 may adjust the cooking steps or even the quantities of the ingredients.

[0105] According to the utility and method descriptions three main processes of using the Al-powered cooking assistant can be described by the sequence diagrams: requesting the personalized recipe 502, requesting the modified recipe 503, and requesting the functional recipe 504. All other processes are trivial and are not described by the sequence diagrams. As already mentioned, the Al-powered cooking assistant comprises multiple devices, however, the user can interact only with the terminal device 300, as it is the only device that contains the display and the user interface 301. Therefore, each process starts with the user interacting with the terminal device 300.

[0106] The process of getting the personalized recipes 502 can be described by the sequence diagram, depicted in FIG.12. The goal of the user is to get the personalized recipe 502 that is associated with the cooking profile 505. In this regard, the user first requests personalized recipes 502 to the terminal device 300. Since the generator method 402 is part of the Al engine 400 the terminal device 300 sends the request to the Al engine 400.

[0107] For the Al engine 400 to generate personalized recipes, it needs additional context information. What are users’s preferences and what are his external environment constraints. It first sends a request for input content to the terminal device 300. Since the terminal device 300 doesn’t currently have that information, it redirects the request to the user by dividing the request into the request for the input mechanism 306 and input content 305. When the user provides the information, the terminal device 300 sends that information back to the Al engine 400. A similar procedure is then done for the adjustment content 307 that is also needed by the Al engine 400 to properly generate the personalized recipes 502. These are then sent back to the user via the terminal device 300.

[0108] The second process of getting the modified recipe 503 can be described by the sequence diagram, depicted in FIG.13. The user starts by modifying the selected personalized recipe 502 or cooking profiles 505. For example, the user’s personalized recipe 502 is vegan lasagna with chestnut mushrooms. However, the user doesn’t have chestnut mushrooms at home and would like to substitute the ingredient with the alternative.

[0109] During the modifications, the terminal device 300 can send suggestions to the user. For example, the terminal device 300 suggest the user to substitute chestnut mushrooms by similar types of mushrooms such as shiitake, button, or oyster mushrooms. When the user is satisfied with the modifications, he sends the request to modify the personalized recipe 502 to the terminal device 300. The terminal device 300 then communicates with the Al engine 400 by sending the request for modification and the modifications themselves. When the Al engine 400 has enough information, it generates modified recipe 503 and sends it back to the user via the terminal device 300.

[0110] The third process of getting the functional recipe 504 can be described by the sequence diagram, as shown in FIG.14. The aim of the user is to get the functional recipe 504 that is associated with the cooking profile 505 and the cooking device 100. The user starts by requesting the functional recipe 504 to the Al engine 400 via the terminal device 300. To properly generate the functional recipe 504, the Al engine 400 requests the terminal device 300 which cooking device 100 will be used. The terminal device 300 first gets the list of cooking devices 100 from the smart home controller 200 and sends the information to the user for selection. When the user selects the appropriate cooking device 100, the terminal device 300 sends the information back to the Al engine 400. Similar procedure is done for acquiring the information which cooking operation 101 will be used. When the Al engine 400 gets all the needed information, generates the functional recipe 504.

[0111] The functional recipe 504 is now optimized in a way that the user and the cooking device 100 can jointly perform all the cooking steps in accordance to the selected cooking operations 101. To perform all the cooking steps efficiently and get to the cooked dish, the functional recipe 504 is sent directly to the smart home controller 200 and not to the terminal device 300 as one would normally assume. This is the consequence of the smart home controller 200 comprising the control method 203 that also controls the cooking steps. This is depicted in FIG.15. It shows the sequence diagram of controlling the cooking steps by the smart home controller 200.

[0112] When starting the cooking steps the smart home controller starts preparing the cooking device 100 according to the functional recipe 504. In parallel it sends requests to perform the cooking steps to the user via the terminal device 300. For example, preparing the ingredients by slicing, chopping, whirring, etc. The user sends the feedback of progress back to the smart home controller 200 via the terminal device 300 to advance over the sequence of cooking steps. When all the cooking steps are finished and the cooking device 100 is prepared, the smart home controller 200 and the user interact with each other via the terminal device 300 to move the food in the cooking device and start the cooking process. During the cooking process the smart home controller 200 can request additional cooking steps to the cooking device 100 and send status updates to the user via the terminal device 300 until the cooking is finished and the dish is ready to eat.

[0113] The best embodiment

[0114] The Al-powered cooking assistant comprises of one cooking device 100, one smart home controller 200, one terminal device 300, and one Al engine 400. Cooking device 100 contains one connectivity module and is managed by one smart home controller via WiFi. The smart home controller 200 functions as an external server. It comprises of one database 201, one controller API 202, and one control method 203. The terminal device 300 functions as a separate device, particularly a mobile device (a phone or a tablet). The terminal device comprises one user interface 301, one input method 302, one adjustment method 303, and one mutually exclusive filter 304. The Al engine 400 functions as an external server. It comprises one Al API 401, one generator method 402, one modification method 403, and one ingestion method 404.

[0115] When acquiring cooking profile 505, only three basic adjustment contents 307 are provided. No additional adjustment contents 307 are required.

[0116] The generator method 402 comprises of one algorithm, in particular multimodal large language model that always generates multiple suggested recipes 502. The same algorithm is used also for the modification method 403 and the ingestion method 404. The modification method 403 has a quick fix functionality with four fix-options, in particular, high protein, vegan, simplify, and gluten free option. The ingestion method 404 requires one selected cooking device 100 and one selected recipe for adjustment.

[0117] All other functionalities stay as they are described in the detailed description.

[0118] Key Differences to Existing Solutions

[0119] 1 . Eliminating the need for specialized devices and data. Our solution is not dependent on a refrigerator or other cooling devices. Furthermore, we don’t need specific measuring devices or medical examination devices that users typically do not possess or find difficult to obtain. We also do not rely on specific data that might be difficult to obtain such as physiological parameters, gas chromatography or other data.

[0120] 2. Comprehensive input space. The system considers the context of the user's personal preferences and their external environment, including cooking experience, current psychophysical state, dietary preferences, health predispositions, and current desires, which is more comprehensive than other solutions that focus only on specific cases within the input space. 3. Eliminating the need for complex recognition systems. Our approach doesn’t rely on recognizing user features from facial expressions or speech. Furthermore, we do not possess systems to extract recipes or assessing cooking ability of the user from video images.

[0121] 4. Elevating simple recipe structures. Our solution doesn’t need special data structures for the recipes or graph connections between the ingredients as it can work with readily available user-friendly recipes, typically found in physical cookbooks. Moreover, it can infer implicit knowledge from the recipes how ingredients relate to each other and what are their implications on the characteristics of the final dish by using a consolidation context from a set of recipes.

[0122] 5. Generative approach. Our solution uses the knowledge of a model to generate the result, making it surely a generative approach, as opposed to other methods that map input data to a vector space and then derive conclusions based on similarity, which is a discriminative approach. This approach provides advanced capabilities of generating new, unseen recipes, allowing users to explore new tastes and cuisines, which is a feature not commonly found in current solutions.

[0123] 6. Advanced Al-driven personalization. Our method includes specific components such as mutually exclusive filter, particular cooking profile structure, consolidation context, modification method with swapping and quick-fix functionality, and specific ingestion method making it superior to other white appliance manufacturers' offerings. Our method searches for the personalized recipe in the input space, as opposed to other solutions that seek for recipes in the target space. Moreover our specific ingestion method is a device-agnostic to device-specific mapping as opposed to other solutions that have device-specific to device-specific mapping.

[0124] 7. Integration with smart home appliances: Unlike other Al-powered recipe apps, our solution offers seamless integration with kitchen appliances or cooking devices, enhancing the capabilities of these appliances for optimal convenience.

[0125] Among all the claims, described in the detailed description, the following are the key technical claims novel to the technical field:

[0126] 1 . Specific cooking profile and its structure. The cooking profile is central to the personalized functionality of the Al-powered cooking assistant. It allows the system to tailor recipes to individual user preferences and their external environment, which is a significant advancement over generic recipe databases or cooking appliances that do not offer such personalization or focus only on specific cases. Furthermore, various mechanisms can create content that can be incompatible between each other or in relation to the methods ingesting them for further processing. The special structure used for the cooking profile is in line with the input space and is comprehensible by methods no matter the content’s form. Random generator. This feature is particularly important as it caters to the spontaneity and variety that users may seek in cooking, especially when they are looking for new ideas. In contrast to other solutions the feature provides recipes without any restrictions. The ability to generate recipes randomly can inspire creativity and help users discover new combinations and flavors they might not have considered otherwise. Furthermore, the random generator comprises a special technique of reel slots, not known in the technical field of personalized recipes, that can be configured to specific ingredient categories and locked if necessary. Adjustment method. The feature directly impacts the personalized recipe recommendation system by allowing for a more tailored result, a significant advancement over other solutions that do not offer such personalization. This method involves a sophisticated algorithm that considers the user's input and converts it to structured data that is comprehensible by all the methods in the Al engine. Mutually exclusive filter. The filter ensures the compatibility and appropriateness of input space feature (i.e., the content) combinations. The mutually exclusive filter checks for discrepancies and corrects them with human-computer interaction, ensuring user's consideration with explicit indication of discrepancies and coherence of the final cooking profile, reflecting the user's true preferences and restrictions. This feature is crucial for maintaining the integrity of personalized recipe generation and avoiding contradictory recipe suggestions, which enhances the final recommendations. Generator method. The generator method's ability to create diverse and customized recipes is a significant technological advancement over existing systems offering a unique solution to the user's culinary needs. It is not merely an Al-powered generative method, as it contains a distinct feature of providing recipes based on the specially designed cooking profile, and consolidation context, utilizing prior knowledge to provide better results. The consolidation context is created by a sophisticated algorithm using basic recipes and additional weight factor given by a probabilistic inference of a Bayesian network. The consolidation context which acts as a prior knowledge how the relation and composition of ingredients and their characteristics affects the characteristics of the final dish. Modification method. The provided feature ensures personalization of recipes to practical circumstances of the users. It comprises of specially designed swapping functionality that can substitute ingredients with similar ones and with a quick-fix functionality that personalizes the recipe with minimal effort. Functionalities, not present in other solutions. Ingestion method. The disclosed feature is an advanced solution providing two key distinctions. To overcome deficiencies of current solutions that provide device-specific to device-specific mapping our method uses device-agnostic to device-specific mapping. Moreover, it allows the user to select a cooking operation among the capabilities of the selected cooking device and convert the whole recipe, not just the cooking parameters, to be suitable for the selection.

Claims

1. CLAIMS1. Al-powered cooking assistant with issue and improve recipe personalization methods and devices where cooking assistant operates in three distinct ways, i.e. generating personalized recipes, exploring new tastes and cuisines, and modifying recipes to available resources wherein it can produce personalized recipes from the input space, adapting to user preferences or the time and energy constraints they have; it can generate new, unseen recipes, allowing users to explore new tastes and cuisine; it can modify recipes to match the available ingredients and the capabilities of the cooking devices.

2. Al-powered cooking assistant, according to claim 1 , have cooking device, smart home controller, terminal device and Al engine, wherein the system have at least one device with a processor and a memory and comprises at least one cooking device (100), at least one smart homecontroller (200), at least one terminal device (300), and at least oneAl engine (400).

3. Al-powered cooking assistant, according to claims 1 and 2, whereinthe smart home controller (200) is a comprehensive device that is configured to manage smart home appliances and comprises at least one database (201), at least one controller application programming interface (API) (202), and at least one control method (203).

4. Al-powered cooking assistant, according to claims 1 and 2, wherein the terminal device (300) can be any device with a display that enables the user of the smart home system to manage it and the terminal device (300) can be integrated within the cooking device (100) or function as a separate device.

5. Al-powered cooking assistant, according to claims 1 and 2, wherein the Al engine (400), comprises at least one Al API (401), at least one generator method (402), at least one of modification method (403) and at least one ingestion method (404).

6. Al-powered cooking assistant, according to claims 1 to 6, wherein the cooking profile is central to the personalized functionality of the Al- powered cooking assistant and allows the system to tailor recipes to individual user preferences and their external environment, which is a significant advancement over generic recipe databases or cooking appliances that do not offer such personalization or focus only on specific cases; furthermore, various mechanisms can create content that can beincompatible between each other or in relation to the methods ingesting them for further processing and the special structure used for the cooking profile is in line with the input space and is comprehensible by methods no matter the content’s form.

7. Al-powered cooking assistant, with issue and improve recipe personalization methods and devices, according to claims 1 to 6, wherein the random generator is particularly important as it caters to the spontaneity and variety that users may seek in cooking, especially when they are looking for new ideas. In contrast to other solutions the feature provides recipes without any restrictions and the ability to generate recipes randomly can inspire creativity and help users discover new combinations and flavors they might not have considered otherwise; furthermore, the random generator comprises a special technique of reel slots, not known in the technical field of personalized recipes, that can be configured to specific ingredient categories and locked if necessary.

8. Al-powered cooking assistant, with issue and improve recipe personalization methods and devices, according to claims 1 to 6, wherein adjustment method directly impacts the personalized recipe recommendation system by allowing for a more tailored result, a significant advancement over other solutions that do not offer such personalization and involves a sophisticated algorithm that considers the user's input andconverts it to structured data that is comprehensible by all the methods in the Al engine.

9. Al-powered cooking assistant, with issue and improve recipe personalization methods and devices, according to claims 1 to 6, wherein the mutually exclusive filter ensures the compatibility and appropriateness of input space feature (i.e., the content) combinations and the mutually exclusive filter checks for discrepancies and corrects them with humancomputer interaction, ensuring user's consideration with explicit indication of discrepancies and coherence of the final cooking profile, reflecting the user's true preferences and restrictions while this feature is crucial for maintaining the integrity of personalized recipe generation and avoiding contradictory recipe suggestions, which enhances the final recommendations.

10. Al-powered cooking assistant, with issue and improve recipe personalization methods and devices, according to claims 1 to 6, wherein the generator method's ability to create diverse and customized recipes is a significant technological advancement over existing systems offering a unique solution to the user's culinary needs. It is not merely an Al-powered generative method, as it contains a distinct feature of providing recipes based on the specially designed cooking profile, and consolidation context, utilizing prior knowledge to provide better results and the consolidation context is created by a sophisticated algorithm using basic recipes andadditional weight factor given by a probabilistic inference of a Bayesian network, in the mean time the consolidation context which acts as a prior knowledge how the relation and composition of ingredients and their characteristics affects the characteristics of the final dish.11 .Al-powered cooking assistant, with issue and improve recipe personalization methods and devices according to claims 1 to 6, wherein the modification method provided feature ensures personalization of recipes to practical circumstances of the users and comprises of specially designed swapping functionality that can substitute ingredients with similar ones and with a quick-fix functionality that personalizes the recipe with minimal effort, which is the functionalities, not present in other solutions.

12. Al-powered cooking assistant, with issue and improve recipe personalization methods and devices, according to claims 1 to 6, wherein the ingestion method is an advanced solution providing two key distinctions and to overcome deficiencies of current solutions that provide device-specific to device-specific mapping our method uses deviceagnostic to device-specific mapping and moreover, it allows the user to select a cooking operation among the capabilities of the selected cooking device and convert the whole recipe, not just the cooking parameters, to be suitable for the selection.

Images