What is Somia Swish? Voice care checks for older people

Somia™ Swish is a voice-based care-check system developed by yeshcube for older people supported in private homes, supported housing, residential care and home-care services. It runs brief scheduled interactions, records whether there is a response and reports the result to authorized relatives, carers or professionals. This voice-based care-check system for older people bounds the conversation to reduce the risk of relational dependency and keeps human intervention within the process. It is at the ideation stage, with no validated technical or care prototype.

The ERL scale (Evidence Readiness Level) measures the evidence maturity of a solution across six levels, from the initial hypothesis (ERL-0) to consolidated impact (ERL-5).

The problem: routine checks and distributed supervision

Everyday support for an older person combines phone calls, visits, in-person rounds and telecare systems. Some of those checks call for direct observation, care tasks or professional assessment. Others only confirm that the person responds, that they heard a reminder or that they want contact.

Four concrete difficulties follow from that. The first is the time absorbed by repetitive checks that almost always end without an incident. The second is telling a missing response apart from a connection failure, a problem with the device or difficulty hearing the question. The third is directing each alert to someone who can actually review it. The fourth is keeping a spoken response from being taken for an assessment of physical, cognitive or emotional condition.

The design principle that answers all four is a brief check that documents what was observed, separates technical incidents and directs human attention toward what needs review.

Design requirements

Five requirements follow from that problem, and they govern the design of Somia Swish:

• Bounded interaction: each check serves one defined care purpose and ends once its result is recorded.
• Human supervision: responses, requests and missing responses reach an authorized person who reads the context.
• Separation of states: a personal result and a connection or operating failure are reported as different things.
• Configurable accessibility: volume, pace, repetitions and waiting time adjust to each profile.
• Proportionate privacy: capture and retention are limited to what running and reviewing the check requires.

Architecture: the device, the supervision layer and the Somia core

Somia’s modular character separates what the core provides from what each product resolves. In that division, Somia Swish is the interface placed beside the person: voice output, capture of the response, execution of the configured schedule, status reporting and connection to the application or the dashboard. It works in a single home and in a multi-device deployment inside a care organization.

The application links profiles, configures the checks, shows the results and manages authorized contacts. The professional dashboard organizes devices by organization, unit, room or shift, and lets an alert be assigned, an action recorded and the follow-up closed.

The core provides the orchestration of the interaction, the authorized content, the safety rules and the data governance, with its scope restricted to the defined greetings, questions, confirmations and reminders. Open conversation and continuous companionship fall outside the architecture.

How a check works

An authorized person configures the schedule, the type of interaction and the review protocol. At the scheduled time, Somia Swish gives a signal, greets the person and asks a brief question or delivers a reminder: how they slept, whether they are awake, whether they need someone to come by.

The person answers by speaking. The system records whether it received a response, whether there is a request for contact and whether the result needs review, and it repeats the interaction when the configured protocol calls for it.

The result reaches the application or the dashboard. A missing response reports that the check was not completed; a connectivity incident appears as a technical failure. The responsible person reviews the alert and records the call, the visit or the action taken.

Care operation: homes, supported housing and care organizations

In a private home, a relative or a carer schedules the checks and receives the result in the application: a confirmed response, a reminder that was heard or a request for a call. Operation requires someone responsible for reviewing the alerts.

In supported housing and home care, a single interface organizes several profiles and the team allocates in-person checks according to what comes in.

In residential and care organizations, checks are configured by person, room, unit or shift. The dashboard shows the completed checks, the requests for assistance, the missing responses and the technical incidents, and staff assign each case and record its resolution. That information directs rounds toward what calls for presence. Rounds for care, observation, professional assessment, environmental checks and social contact remain in person.

The bounded interaction also protects the purpose of the system. Somia Swish greets, asks, reminds, confirms and offers human contact, and the exchange ends once the action is complete. The system does not simulate affection, does not present itself as a carer and does not provide continuous emotional companionship.

Data and privacy

The person knows when a check takes place, what is recorded and who can consult it. The system handles schedules, responses, requests, alerts, technical incidents and follow-up actions; in a care organization, traceability also covers the execution of the check, the professional assigned and the closure of the case.

Somia’s personalization memories fall outside the functional scope of Swish: the checks do not build a conversational profile of the person.

Processing is governed by the GDPR and the Spanish LOPDGDD, with Yeshcube Tech, SL as data controller. The applicable framework is published in the Somia privacy policy and in the AI transparency commitment. What the product retains on its own account — audio, transcript or structured state, with its retention period, its purpose and the access rules of a care organization — will be declared with its technical and legal specification.

Artificial intelligence models read a response incompletely, classify it wrongly or miss the context, and an uncertain result needs human review. The design also accounts for the accidental capture of other people’s voices.

Evidence and validation status

Somia Swish holds level ERL-0, initial hypothesis: the proactive voice check, the limit on conversation, the alert to whoever supervises, the professional dashboard and the separation between a missing response and a technical incident are defined, with no evaluated prototype, sample, test context or measurements of its own.

Still to be determined are the device and its indicators, the validated repertoire of questions and reminders, the states of a check, the number of attempts and the waiting times, the classification and escalation criteria, the scope of the application and the dashboard with their roles and permissions, the data retention regime and the criteria for use with people living with cognitive impairment. Each one is published with the specification that settles it.

Level ERL-1 (proof of concept) requires a prototype that allows evaluation of comprehension, voice capture, result classification, false alerts, accessibility, dashboard operation and follow-up management. In a care organization it must also measure the effect on rounds, the alert workload, staff time and the quality of human contact. It will be declared once that evidence is available.

Commercial transfer is conditioned by the hub’s general rule, which requires at least ERL-3 to transfer any solution.

What it is for and what is left out

Somia Swish is intended for everyday checks in private homes, supported housing, residential care, care organizations and home-care services: confirming that a person responds, delivering reminders, recording a request for contact and helping to prioritize in-person checks.

The system offers non-clinical care support. It does not make diagnoses and it does not detect falls, loss of consciousness, physiological changes, cognitive decline, actual medication adherence or emergencies. A response received does not establish the person’s overall condition, and a missing response calls for human verification. Left out of its definition is replacing professional care, care tasks, clinical rounds, emergency services or human contact.

Somia Swish takes the place of the bounded care check in the ecosystem, with the application and the dashboard as the supervision points. Its certification under Somia™ Within is conditioned on technical integration. The details of the system behind it are on the Somia site.