Datacom LLC is a pioneering Mongolian tech company that serves as the sole registrar of .MN domain names and provides digital business solutions. Founded in 1994, the company introduced the Internet to Mongolia in 1996, enabling people to connect globally and access information like never before. From that point on, the company gained greater recognition, and many businesses began relying on Datacom LLC for their services.

On average, we handle 100 to 300 customer support requests daily via calls, chats, and emails. Most inquiries involve frequently asked questions, such as office email setup, domain DNS settings, or Google Workspace features. While these guides are available in our public knowledge base, customers rarely access them. I realized we could develop an AI assistant to first analyze and respond to customer questions accurately. If the AI cannot provide an answer, it automatically transfers the request to the appropriate department, where a staff member can assist. This approach significantly improves the efficiency and quality of our customer support.

I needed an AI model capable of understanding Mongolian and providing grammatically correct responses. After extensive research and comparing all available models, I found that OpenAI’s ChatGPT 4o was the most accurate for Mongolian. It can understand input in both Cyrillic Mongolian and Mongolian written with Latin letters, making it a perfect fit for our needs.

DataChat, the main platform

Before developing and training the AI model, I needed a user-friendly platform where customers could chat with both the AI and our staff. This led to the deployment of a platform called DataChat. It uses WebSocket for real-time message communication, ensuring instant message delivery. The frontend was built with Next.js and Shadcn, featuring a minimal and intuitive UI design.

Using this platform, staff members can efficiently manage customer interactions and resolve issues promptly.

After building the platform, I began training the AI model using our knowledge base so it could provide answers based on our service details. I named it Daco AI, with "Daco" being short for Datacom. I also created a small, cute robot illustration as the AI’s avatar.

Daco AI has significantly improved our customer support by handling 60–70% of requests independently, and customers have responded positively to its assistance. After implementing this feature, support calls decreased, allowing staff members more time to focus on personal development and improvement.