We are building a multilingual artificial intelligence voice agent that will automate customer service and planning at companies that provide periodic servicing.
The product addresses a problem common in service companies: customer service is inefficient because a large share of staff time goes on repetitive phone calls, giving information, agreeing visits, registering orders and other standard tasks. With limited staff, customers often face long waits, missed calls or uneven service quality.
In real time, the system will tell whether a negative answer comes from an objection that can be overcome or from circumstances in which continuing the offer no longer makes sense. This will let it choose the right action: give another argument, offer an alternative, postpone the contact or end the conversation respectfully. This way the agent will avoid pushy selling, or selling that does not fit the customer's situation.
The agent will speak Lithuanian, Russian and Polish and understand English technical terms. When a customer switches language mid-sentence, the agent will follow, with no manual selection.
It is intended for companies that provide periodic servicing in water treatment, heating and ventilation, household and medical equipment, fire protection systems and lift maintenance. The solution will be easy to adapt to service companies in various fields, so it can be replicated and deployed in different business sectors without major technological changes.
It will be delivered as a service (SaaS), with a web console for the company. For the end customer, the phone call itself will be the interface.
The agent will be built from eight layers: telephony, streaming speech recognition, prosody analysis (intonation, pauses and speaking rate), a dialogue policy, answers from the company's knowledge base, speech synthesis, integrations with the company's systems, and compliance. Planned technologies: Python and PyTorch, open-source speech recognition and synthesis models running on infrastructure we control, and SIP/VoIP telephony.
Telephony
SIP / VoIP: incoming and outgoing calls
Done by AI: Speech recognition
Streaming, without waiting for the speaker to finish
Done by AI: Audio and prosody analysis
Pauses, speaking rate, intonation, voice energy
Done by AI: Dialogue management and conversation policy
Choosing the next step, classifying the reasons for a refusal
Done by AI: Building answers from knowledge
Based on the company's knowledge base: services, prices, working hours
Done by AI: Speech synthesis
To be adapted to Lithuanian stress and domain terms
Integrations
REST API and MCP-type interfaces to the company's systems
Compliance and control
Consents, do-not-call lists, audit trail
The call
Done by AI: Analysing the call and its prosody
Getting the actual outcome from the company's system
Done by AI: Updating the model's behaviour
A better decision in later calls
Next call
A replicable product that adapts to each organisation.
We plan to use our own and rented AI infrastructure, operating in the European Union region.
EU AI Act: at the start of each conversation, the person will be told clearly that they are talking to an artificial intelligence system.
This is where we start: the agent will agree a periodic service visit with the customer and create the task in the company's system.
Calls in which the agent will explain why maintenance is needed when this isn't obvious to the customer.
Calls to customers who have stopped using the company's services.
A pilot at a company that provides periodic servicing.
Validating the product on the results of the pilot.
Rollout to other periodic-service companies in the sectors listed above.
Commercialisation is based on a standardised adaptation pipeline, delivered as a service. For each new client, only the last stage is tailored individually: adapting the model to that company's conversation data and business processes.
Our experience closest to this product is Langer: a messaging and calling app that translates text and speech in real time.
2026Beta
A messaging and calling app with a built-in interpreter: chats and calls are translated as you talk, so everyone can speak their own language.
The working prototype already translates chats, voice calls and face-to-face conversations, and beta sign-ups are open.
iOS and Android, speech recognition and synthesis, AI translation, real-time back end
If your company provides periodic servicing and you would like to know more about the agent we are building, send us an enquiry.