Back
Key Takeaway
Achieved 95% answer accuracy and established a corporate knowledge utilization system through prompt tuning tailored to data characteristics
Through continuous feedback-based enhancement and final prompt tuning matched to data characteristics, established an initial sales data utilization system and achieved improved answer accuracy up to 95%
Hansol Paper (한솔제지)
Client :Hansol Paper (한솔제지)
Industry :Manufacturing
Service Area :Data & AI
Applied Solution :AIR
1. Overview (Project Background)
Hansol Paper initiated this project from the necessity to convert years of accumulated sales logs into a database, discover hidden insights within it, and actively utilize them in business. The main motivation is to uncover valuable information from existing records, improve work efficiency, and create new business opportunities.
Project Goals
Establish a corporate knowledge utilization system based on generative AI utilization
Enhance accessibility to historical information needed for task execution
Form a Shadow-IT control culture through internalization of generative AI utilization
2. Solution (Resolution Approach)
We validated an AWS-based platform to identify the most suitable resources for Hansol Paper's requirements.
Components
Bedrock, Claude: LLM model capable of maintaining conversation flow and answering questions
Bedrock(Agent): AI Agent development tailored to specific topics or services
AmazonQ: AI assistant service providing real-time voice responses to Amazon-related service inquiries
AmazonOpenSearch: Easily build and operate search engines in cloud environments
3. Result (Achievements)
Established a system (platform) enabling corporate knowledge utilization
Improved answer quality through generative AI to 95% through prompt tuning tailored to data characteristics
Continuous quality enhancement based on generative AI
Enhanced data processing
Enhanced synonym/similar word dictionary
Date processing
Enhanced query performance based on vector similarity
