Toss Improves Developer Documentation Accessibility with RAG-based Chatbot 'Mr. Park' and Documentation Automation Tool 'Sillokbot'

The Toss Frontend Chapter built a system where documentation comes to the developer, instead of users having to hunt for documents, to solve the problem of developers struggling to find internal tool documentation. To overcome the limitations of existing documentation structured from the author's perspective, which caused time-consuming searches, they observed developers' natural workflow and reflected it in the design.

RAG-based Chatbot 'Mr. Park'

They developed Mr. Park, a RAG (Retrieval-Augmented Generation)-based chatbot, to provide immediate answers as if developers were asking colleagues. Mr. Park works within IDEs like VSCode and Cursor, as well as internal messengers, providing accurate answers based on existing documentation along with sources. This enables information retrieval through 'conversation' rather than 'navigation', delivering more accurate and consistent information than colleagues.

Documentation Automation Tool 'Sillokbot'

To maintain a high-quality knowledge base, they introduced Sillokbot, a tool that automatically documents internal messenger conversations. By reacting with emojis or calling the bot in a thread, AI analyzes and summarizes the conversation to generate a PR (Pull Request). This automatically captures knowledge during the problem-solving process, eliminating the burden of writing separate documentation, and the generated documents are used as training data for Mr. Park.

Results and Future Plans

The introduction of the two tools ensured that knowledge flows naturally across the entire team rather than being stuck with specific individuals, reducing duplicate questions and improving team productivity and efficiency. Toss plans to continuously improve its learning infrastructure by advancing its AI-based documentation system, automating documentation linked to code, and strengthening integration with internal platforms in the future.