What makes Somia a modular conversational AI system?

Somia is the modular conversational artificial intelligence system developed by yeshcube for everyday emotional wellbeing experiences. Its architecture brings orchestration, content, data, analysis and validation together in a shared core that can be integrated into physical and digital products with different interfaces, sound systems and contexts of use.

The modular structure allows an acoustic booth, a portable mask or a home device to share interaction rules, safety criteria and data-governance mechanisms. Each product retains its own layer for physical form, controls, connectivity and environment of use.

The problem modularity addresses

Emotional language contains metaphors, ambiguity, venting and expressions whose meaning depends on context. A general conversational model may interpret these signals only partially, respond inappropriately or continue a conversation outside its intended scope.

Deployment across different products creates another difficulty. An experience in an acoustic booth presents different conditions from a session using a portable mask or home device. The microphone, audio system, privacy of the space, duration, activation method and available controls all change.

Somia addresses both problems through a shared architecture. The core retains the common rules and each integration defines the capabilities that belong to its product.

Five shared layers

Somia’s architecture is organized into five separate, coordinated layers.

Layer	Function
Orchestration	Decides which experience is activated, its sequence and the applicable rules
Content	Brings together sound experiences, guided conversations and available exercises
Data	Manages personal memories, permissions and usage signals
Analysis	Turns aggregated information into measurements for evaluating and improving solutions
Validation	Applies the evidence, safety and integration criteria for each deployment

This separation allows content, providers or interfaces to change without redefining the entire architecture. It also distinguishes shared capabilities from product-specific functions.

The core provides the conversational models, layered instructions, safety rules, content and measurement mechanisms. The product layer handles audio capture and playback, physical controls, the configuration app, connectivity and industrial design.

How an interaction works

A conversational interaction begins when the person voluntarily activates the microphone available in the product or on their personal device. The audio is processed through the voice infrastructure and converted into a transcript the system can interpret.

The layered instruction system analyzes the context of the expression. It distinguishes factual information, metaphorical language, venting and possible risk signals. The result determines whether the system continues the conversation, adjusts the experience or applies a stopping rule.

The response is generated within the scope of non-clinical wellbeing and played through speech. The person should know they are interacting with artificial intelligence and retain the ability to pause, end or change the configuration.

yeshcube uses the term algorithmic empathy for this capacity to interpret contextual signals and adjust conversational behavior. The term describes a technical process and carries no emotional meaning for the machine.

Interpretation and risk-detection mechanisms have limitations. They may miss nuance, produce false positives or fail to recognize a relevant situation. Somia does not provide clinical intervention or contact emergency services.

Two experience modes

Somia organizes its experiences into two main modes.

Somia Sensory

Somia Sensory brings together predefined sound content for breathing, relaxation, concentration or rest. The sequence is established before the session begins and works without maintaining a conversation with the system.

This mode can use voice, music, soundscapes and silence. Its availability and content depend on each product.

Somia Immersive

Somia Immersive adds the conversational layer. The person can describe their context, select an objective or respond to the experience’s guidance. Somia adapts the language and sequence to the available information and authorized preferences.

Adaptation is based on the content of the interaction and the data the person has chosen to share. It does not amount to a clinical measurement or an infallible reading of emotional state.

Products and deployment layers

The modular architecture takes shape in products that address different contexts and retain their own development status.

Product	Specific layer	Documented status
Somia Bloom	Spatial sound and integration into acoustic booths	ERL-3, initial transfer
VVAVVE	Portable mask with immersive audio and gentle light reduction	ERL-2, controlled pilot
Somia One	Personal home device using voice and sound	ERL-0, initial hypothesis

Somia Bloom provides the physical audio installation for acoustic booths. Its Echo and Pulse systems adapt to different volumes and installation modes. The booth has no integrated microphones and uses a mobile phone as its control point and voice input when required by the selected mode.

VVAVVE moves the experience into a portable format. The mask combines immersive audio, reduced visual stimulation and bounded sessions. Its evidence level applies to the evaluated prototype and does not establish clinical efficacy.

Somia One is planned as a home access point connected to Somia App. Its purpose, interaction and relationship with the core are defined, while industrial design, technical components and commercial availability remain for later stages.

The status of each product keeps the maturity of the shared architecture separate from the evidence for a specific integration. Adding the Somia core to a new device does not automatically transfer the results obtained with another format.

Infrastructure connecting the ecosystem

Modularity also applies to the infrastructure supporting deployment.

Somia Within

Somia Within establishes shared technology, ethics and evidence criteria for first-party and third-party products. Its role is to check that an integration retains the system’s rules and documents the responsibilities of each party.

The certification establishes the evaluated scope. Wellbeing outcomes, safety in each context and the efficacy of an experience require specific evidence.

Somia Atlas

Somia Atlas is the ecosystem’s territorial layer. Its purpose is to organize the discovery of available spaces, products and experiences according to location and access conditions.

Somia Data

Somia Data is the measurement and intelligence infrastructure under development. Its purpose is to turn aggregated information on use, continuity and utility into dashboards and reports for improving solutions.

Somia Data is at ERL-0 and its technical architecture has yet to be specified. Its principles exclude the sale of personal data, advertising profiling and the exposure of individual conversations in reports for organizations.

Data governance

Somia separates aggregated data from personal memories.

Personalization memories are activated through a decision by the person. They should be available for viewing, editing and deletion. Their use depends on the product, account, permissions and available configuration.

The architecture takes a local-first approach to personal memories. Information remains on the device by default and is synchronized when the person activates that option. Each product should document which data it retains locally, which information it transmits and which providers are involved.

Processing is subject to the GDPR and Spanish data-protection law. Data protection by design requires collection, retention and access to be limited to what is necessary for each purpose.

Voice interaction creates specific requirements. The product should communicate when the microphone is active, how the audio is used, whether a transcript is created and how the person can exercise their rights.

Transparency and human control

Somia keeps its technological nature explicit. The voice and conversational style may feel natural, but the system should not present itself as a person or claim emotions of its own.

The European Union Artificial Intelligence Act establishes transparency obligations for systems intended to interact directly with people. This requirement aligns with Somia’s principle of disclosing the artificial nature of the interaction.

Decisions with significant consequences need additional controls. Somia is intended for everyday wellbeing and its responses should not be used as diagnoses, prescriptions or professional judgments.

Human oversight also applies during development. Changes to models, instructions, providers, data or products need testing for regressions and new risks.

Evidence and maturity

Somia holds ERL-3, initial transfer. This level reflects the integration of the architecture into first-party and third-party products and experience of deployment in real settings.

The next level, ERL-4, requires evidence of scaling across several contexts and independent review. Somia will declare it when studies and results meeting those requirements are available.

The system’s level does not make every feature equally validated. Each product, mode and use needs its own evaluation. Conversation, sound, hardware and context may influence safety, acceptance and outcomes.

Intended scope

Somia is intended for the everyday emotional wellbeing of adults. Its applications include relaxation, concentration, self-care and conversational support within a non-clinical scope.

The following remain outside its scope:

• Diagnosing disorders or diseases.
• Prescribing treatment.
• Replacing healthcare or psychological professionals.
• Emergency response.
• Guaranteeing correct interpretation of every risk-related expression.
• Clinical inference of emotions from voice.

In a crisis or situation involving possible harm, the appropriate route is an emergency service or qualified professional.

The value of a modular architecture

Modularity allows different products to be developed on a shared core while maintaining a clear distribution of responsibilities.

yeshcube governs the conversational architecture, content, safety rules, data and validation. Each product defines its interface, hardware, installation conditions and context of use. External manufacturers retain responsibility for their own devices.

This structure supports transfer, updating and independent evaluation of each component. It also keeps the limitations of each integration visible and avoids presenting the entire ecosystem as a single solution with one uniform level of evidence.

Documents and references

• Somia, AI Transparency Commitment
• Somia, AI User Agreement
• Somia, Privacy Policy
• NIST, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile
• World Health Organization, Ethics and governance of artificial intelligence for health
• European Union, Artificial Intelligence Act
• European Union, General Data Protection Regulation

Visit the Somia website →

Discover our Allies program →

Fill in our contact form →