At DialOnce, artificial intelligence (AI) makes it possible to understand and guide every customer request with precision.
Our conversational engines are built on a hybrid architecture that combines Natural Language Understanding (NLU) with the most advanced Large Language Models (LLMs) available on the market.
This proprietary approach, which we call Blend, merges the rigor of NLU, used to detect intent and emotion, with the flexibility of LLMs, which generate natural, accurate, and contextualized responses.
The main area of development at DialOnce lies in the transition to a multi-agent architecture, built on two pillars: A2A (Agent-to-Agent) and MCP (Model Context Protocol).
A multi-agent system enables multiple autonomous, specialized agents to collaborate and exchange information in order to accomplish complex tasks together.
Thanks to the A2A protocol, these agents can communicate with each other or with third-party systems (CRM, chatbots, business tools) without any disruption in the user journey.
The MCP protocol ensures that such exchanges take place without exposing sensitive data, as each agent only accesses the information strictly necessary for its task.
This evolution allows each stage of a conversation to be handled by a dedicated agent, whether it is identifying intent, processing a request, or activating a business tool. For example, during the creation of a customer service ticket, the conversation can be automatically routed to the specialized agent in charge, ensuring both precision and continuity.
This architecture will enhance the relevance and reliability of interactions, while laying the foundation for an ecosystem of interoperable, secure, and trustworthy AI.
DialOnce’s AI is built on three essential principles: transparency, human oversight, and compliance.
Our solution incorporates the concept of “LLM as a Judge”: an AI agent configured as a reviewer, responsible for evaluating the quality of interactions a posteriori based on key indicators such as resolution rate, user satisfaction, and answer accuracy.
This agent regularly reviews a representative sample of conversations (around 10%) to identify those that did not meet the expected outcomes or generated user dissatisfaction.
But its role goes beyond analysis: it actively contributes to the continuous improvement of the solution. Based on observed trends, it generates automated recommendations to optimize conversational flows, response phrasing, and routing logic.
The insights collected then feed into a joint improvement loop, shared between our teams and our clients’ teams during feedback sessions and steering committees.
This collaborative process ensures an AI that is evolving and contextualized, able to adapt to real-world usage, environments, and business objectives.
