Hanwha Ocean Adopts AWS-Based AI Translation App OTalk to Bridge Communication Gaps in Shipbuilding Sites

Hanwha Ocean has adopted OTalk, an AI translation app built on AWS cloud and powered by Flitto's multilingual communication service Vogl, to resolve communication gaps among workers of various nationalities within its shipyards. The service recognizes voice even in noisy environments and uses Amazon Bedrock to refine translation results to match shipbuilding business contexts and terminology.

Architecture and Technical Implementation

OTalk processes requests using Amazon CloudFront, ALB, and Amazon EC2, while speech-to-text (STT) and machine translation (MT) are performed on dedicated GPU EC2 instances. To improve translation accuracy, the system uses the Converse API of Amazon Bedrock to refine translations into structured output (JSON schema) based on the client's glossary and source text. Data is stored in Amazon RDS, ElastiCache, and S3, with infrastructure isolated and data security enhanced through a dedicated VPC and private subnets.

Translation Quality Improvement and Field Application

Flitto collaborated with Hanwha Ocean and Hanwha Systems to collect voice data containing shipbuilding terminology, dialects, and field noise for model training. During field validation, the solution was selected with priority on speech recognition accuracy and translation readability, and evaluation data was separated from training data to verify model processing quality.

Adoption Impact and Expansion

As of early October 2026, the usage rate of OTalk among direct foreign personnel reached approximately 96%. Initially applied primarily to direct foreign personnel, the user base expanded by approximately 4.5 times to include the entire company and partners as field usage experience spread. The scope of use expanded to various business communication channels, including work instructions, safety training (TBM), and shipowner meetings, and accessibility was improved by integrating with internal systems to allow login via employee ID without separate accounts.