Lawfirm Develops 'Minbungi' VoC Automated Investigation AI Agent at AWS-Supported Hackathon
Key point
The agent, deployed on the day of the hackathon, aims to reduce VoC tasks that previously took 39 hours per month down to 8 hours.
Details
Lawfirm hosted an internal AI hackathon with support from AWS and unveiled the VoC automated investigation agent 'Minbungi' developed by the Platform Squad 'Plabang' team. This agent was deployed on the day of the hackathon and immediately integrated into actual customer complaint (VoC) handling workflows.
Resolving VoC Processing Bottlenecks
Lawfirm introduced the agent to improve a situation where an average of 29 VoCs were received per month with a median processing time of approximately 80 minutes per case. Previously, inefficiencies occurred in searching for similar cases, gathering basic information, and understanding code flow. 'Minbungi' is designed with a collaborative structure where AI handles information gathering and draft creation, while humans perform verification and make final decisions.
Architecture and Technical Implementation
The agent utilizes the query function of the Claude Agent SDK to handle tool-calling loops and runs on a single server without a separate orchestration server. GitHub and Notion are connected via official MCP servers, while a Custom MCP server was developed for Slack to include only necessary features, reducing authentication complexity. Investigation context is stored in Slack channels instead of a separate database to streamline operations.
Security and Operational Strategy
Considering the characteristics of legal services, three layers of safety measures were applied: blocking local file access, managing GitHub repository whitelists, and strengthening system prompts to prevent indirect prompt injection. Additionally, partial results are returned when API call limits are exceeded, and confidence levels are indicated in responses to minimize hallucinations, encouraging human final verification.
Expected Impact
With the introduction of 'Minbungi', it is expected that communication between handlers will be reduced to zero, and the time from intake to response will be shortened from an average of 2 days to just a few minutes. Lawfirm aims to reduce investigation time per case by 80% and cut monthly VoC response labor from 39 hours to around 8 hours.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.