Innovative Personalized AI Model




CANDiY AI enhance personalization in three different ways

to create hyper-personalized AI models.


Personalized Foundation Models 

We employ innovative class-wise federated averaging during foundation model creation, generating multiple models for each class. This ensures each device's unique characteristics are integrated, creating truly personalized foundation models.

Retraining Device Data



When deploying personalized foundation models to devices, we retrain them using the device's own data and feed the results back. This further personalizes the lightweight AI models applied to devices.


Sharing AI Parameters



As device data changes over time, we innovatively share AI parameters among users with similar profiles to update the personalized models. This allows us to understand the user's evolving situation and context more precisely.



Creating Personalized Foundation Models

During the foundation model creation phase, when transmitting the AI parameters from devices to the cloud, we introduce an innovative class-wise federated averaging technique to generate multiple foundation models corresponding to each class. This groundbreaking approach ensures that the unique characteristics of each device are seamlessly integrated into the foundation models, enabling the creation of truly personalized foundation models.


Retraing Device Data during Lightweight AI Models

In the process of compressing and deploying the personalized foundation models to devices, we implement a novel method where the models are retrained using the device's own data, and the results are fed back into the foundation model. This innovative approach further personalizes the lightweight AI models applied to devices, enhancing their adaptability and effectiveness to meet individual user needs.


Sharing AI Parameters among User with Similar Profiles

After deploying these highly personalized AI models to devices, we recognize that data distribution on devices can change over time. To address this, we have developed an innovative strategy of sharing AI parameters among users with similar profiles to update the personalized AI models. This allows us to understand and adapt to the user's evolving situation and context with greater precision, continuously refining the AI experience.

User-Centric Privacy Design




Through the dual privacy settings and privacy report features, users can transparently monitor how their data is being utilized, and they have the option to immediately revoke data usage approval directly from the interface.


Dual Privacy Control

We provide manual settings feature allowing users to control data collection, storage, usage, and deletion on a per-folder or per-data basis. Our automatic settings detect personal information types, letting users specify AI learning usage for each. Combined, these features prevent unintended data use in AI.

Interactive Privacy Report

CANDiY's interactive Privacy Report visualizes real-time user data usage for AI learning. Users can monitor data and instantly revoke permissions directly within the report, eliminating the hassle of switching screens, enhancing security and user experience.

Personal Vault

All personal data and AI models in CANDiY are securely stored in a personal vault, accessible only with the individual's encryption key, blocking unauthorized access. Our vault works across environments—on-device, cloud, and on-premise—meeting diverse market needs and setting new cybersecurity standards.


Manual and Automatic Settings

CANDiY AI offer an innovative manual settings feature that empowers users to control the collection, storage, usage, and deletion of their own data on a per-folder or per-data basis. Additionally, solution includes an automatic settings function that intelligently detects types of personal information and allows users to specify whether each type can be used for AI learning. These two features complement and integrate with each other to prevent any unintended data from being utilized in AI processes


Real-Time Interactive Privacy Report 

CANDiY AI introduces a groundbreaking approach by visualizing the entire cycle of user data collection, storage, usage, and deletion for AI learning through a real-time, interactive Privacy Report. Unlike simple document formats, this report, based-on interactive web page, allows users to transparently monitor data usage in real time at the moment in the interface. Users can also instantly revoke usage permissions directly within the report. This innovative feature frees users from the hassle of checking statuses and changing settings across different screens, enhancing both security and user experience.


Secure Personal Vault

All personal information and AI models used in CANDiY AI are securely stored within a personal vault. Accessible only with an encryption key issued exclusively to the individual, this vault effectively blocks unauthorized external access, representing a significant advancement in data protection. Furthermore, our personal vault can be applied across various environments—including on-device, cloud, and on-premise servers—addressing diverse market requirements and setting new standards for cybersecurity in AI applications.

Seamless Adaptability and Scalability





CANDiY AI ensures easily integrated and customized across diverse devices, also allows businesses to deploy AI solutions on a wide scale, efficiently and cost-effectively.


Modular Components

The modular components can be applied individually or integrated, allowing for flexible adaptation to various service environments and requirements.


Eco-friendly Efficiency

CANDiY AI does not transmit personal information over the network, which reduces the costs of technical measures required for regulatory compliance and minimizes the use of cloud infrastructure.


Modular Components

The modular components include user data collection, encrypted data storage, and AI usage consent management. Each module can be applied individually or integrated, allowing for flexible adaptation to various service environments and requirements. This modularity enables service providers to quickly and easily establish on-device personalized AI services without the need for extensive infrastructure.


Eco-friendly Efficiency

CANDiY AI completely eliminates the need to transmit personal information over the network, which significantly reduces the costs associated with implementing technical measures required for regulatory compliance. By ensuring that all personal data processing occurs locally on the device, we minimize the reliance on cloud infrastructure, leading to substantial decreases in operational and maintenance expenses.







What makes CANDiY AI special?


CANDiY AI is human-oriented rather than technology-oriented. To achieve technical perfection, we have data privacy experts, PhD-level AI leaders, and a CTO specialized in IT architecture. Additionally, to study the interaction between people and technology, we have a PhD in psychology and a designer who majored in Media Studies. CANDiY AI researches technology for people, not technology for technology's sake.