Kakao Developer Who Built KDCA Chatbot: "Emergency Deployment via k8s... Prototype in One Day"

A backend developer at Kakao shared their experience developing a COVID-19 response chatbot at the request of the Korea Disease Control and Prevention Agency (KDCA). The initial KDCA channel only supported 1:1 inquiries, making communication difficult, but the developer built the backend using Kotlin and Spring Boot and established an automated deployment system in a Kubernetes (k8s) environment utilizing Argo CD. The basic functionality was completed in one day, and the completion level was approximately 85% at the time of launch two days later.

Key features include 'Find Clinic', which accepts location information, integrates address search APIs and place search APIs, and applies logic to convert WGS84 coordinates to the WCONG coordinate system for map image generation. The 'Status' feature queries posts with specific patterns ('Status', 'Epidemiological Investigation') from the KDCA Talk Channel via an internal API and provides them in card format; this was the most frequently used feature. Additionally, since the default knowledge + threshold (0.65) of Open Builder was low, potentially leading to incorrect answers, the developer customized the logic to determine similarity scores independently, thereby improving accuracy.

After the service launched, push messages were sent to 17 million people, and the number of friends on the KDCA channel surged from 300,000 to over 700,000. During traffic spikes, the number of replicas in k8s was adjusted for flexible response. While call volumes were currently on a downward trend, they have returned to an upward trajectory.