The goal was to build an automated front desk solution to deliver a highly personalized user experience to site visitors and boost conversion rates. The solution aimed to provide relevant and accurate information about the company to various types of visitors, including potential clients, candidates, and researchers.

LangChain was used to abstract the specific model and cloud, allowing for flexible switching between models and avoiding dependency on a particular cloud provider. Simple tasks could be delegated to basic models, reducing the cost of their execution. What’s more, this solution can be deployed as a pre-built solution in various clouds, avoiding vendor lock-in.

BigQuery was used as a data warehouse for storing the corporate database.

The Site Assistant consists of a Supervisor and four types of topic-related assistants:

• Supervisor: Analyzes incoming messages from users and decides which type of assistant will handle them.
• Potential Clients Assistant: Addresses questions from potential clients.
• Potential Candidates Assistant: Handles inquiries from job candidates.
• Researchers Assistant: Responds to questions from researchers.
• Technical Support Assistant: Provides technical support information.

If a user wishes to communicate with a real person, the Assistant will provide the appropriate contact information. If the message is not related to Zoolatech, the Assistant will provide a standard fallback message to inform the user that their question cannot be answered.

The chat window is integrated into the site as a widget, accessible throughout the site.

A Site Data Scraper fetches information from the Zoolatech site, processes it, and stores it in a vector-type database.

As a result, a flexible solution was achieved in terms of cost (optimum cost-value) and adaptability to different types of cloud environments depending on the client’s needs. It is used as an in-house solution and is also offered to clients as a pre-built solution that can be easily integrated into their own business ecosystem and customized to their specific needs.

The solution helps automate routine front desk tasks and reduces the load on human receptionists by approximately 60%.