Page 438 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

Use Case 9: Developing an AI-powered translation app to improve

workplace safety at construction and shipbuilding worksites

Organization: Crowdworks AI

Country: South Korea

Contact Person(s):

Sue Kim, hs�kim@ crowdworks �ai
Jung ho KIM, kimjuh@ crowdworks �ai
Jeong Yeon KIM, waymaker@ crowdworks �ai

1 Use Case Summary Table

Item Details

Category Productivity

South Korea’s construction industry employs over 100,000 foreign
workers, a number that continues to grow. However, existing transla-
tion and interpretation tools are often inaccurate and fail to recognize
Problem Addressed
the slang and terminologies commonly used on construction sites—
leading to communication breakdowns and increasing the risk of
serious accidents.
The solution enables seamless real-time communication between
workers at construction and shipbuilding worksites. Unlike the trans-
lation services currently available at the market, Crowdworks AI’s
solution accurate translates industry jargon, thereby preventing
safety hazards. The solution utilizes below technologies:
Key Aspects of Solution
• Multilingual corpus collection for construction and shipbuilding
domain-specific translation model training
• Fine-tuning Gemma-2-9B-IT model [3]
• Applying PEFT via LoRA for efficient model training
• In-context learning and prompt engineering

Technology Keywords LLM, Building Domain-Specific Datasets

Public (Glossary of National Construction Standards [1] by the Minis-
Data Availability try of Land, Infrastructure and Transport (MOLIT)) + Private data from
Korean construction sites
Metadata (Type of Data) TEXT

Model Training and Fine-tuning, Prompt-Engineering
Fine-Tuning

402