講演情報
[5O3-IS-5b-06]Agile In-House Development of a Risk Assessment AI Application for Field Surveys
HIROKI SHIBUYA1, 〇Khaing Wint Zwe2, Gilbert Kurniawan Hariyanto2 (1. PACIFIC CONSULTANTS CO., LTD., 2. Human Resocia Co., Ltd)
work-in-progress
キーワード：
knowledge management、knowledge sharing、knowledge use/reuse
This research developed an in-house RAG-based AI application that utilizes past accident cases and internal risk assessment data to support accurate safety risk evaluation for construction consulting work. It addresses challenges faced by inexperienced young engineers conducting safety assessments before field surveys in diverse environments including steep slopes, rivers, ports, harbors, water and sewerage facilities, and underground spaces. A cloud architecture using Azure OpenAI, Azure AI Search, and Azure Document Intelligence was adopted to leverage company-accumulated knowledge assets effectively. Development proceeded through agile processes in three stages: understanding limitations of general-purpose large language models, introducing standardized risk assessment templates with past accident cases, and utilizing historical risk assessment data through parallel search strategies. As a result, the system achieved practical-level performance, enabling extraction of valid safety risks and receiving positive user evaluations for operational usefulness. Future challenges include integrating external accident databases, improving usability, and enhancing robustness for irregular situations.