Vazco was brought in to develop a solution that would optimize Drukomat's customer support operations and improve the overall user journey on their website. Our goal was to build an AI assistant capable of handling the majority of simple customer inquiries and guiding users through the complex ordering process.
Instead of deploying a basic model that can be prone to "hallucinations" and difficult to control, we designed and implemented an agentic architecture built on our proprietary Nautilus platform. This approach creates a structured workflow where the assistant operates like a team of specialized agents, each with a specific role and access to carefully curated knowledge and tools.
The workflow itself is comprised of multiple steps, and the agents guide the user through this process, semantically evaluating the situation at each stage to ensure all business-defined conditions are met before moving forward. This semantic evaluation – a soft, AI-driven assessment rather than a rigid, data-based check – is a core capability of the Nautilus platform, setting it apart from other frameworks and enabling a more human-like, flexible interaction. The entire process was executed with a lean delivery methodology, focusing on fast, iterative development and continuous feedback.
