JapanConstruction and engineering2024 – ongoingDai-Dan
In-house generative AI chat with search over internal regulations and technical documents
Dai-Dan raised answer accuracy on chart-heavy documents to 80% with an annotation tool
Problem
Dai-Dan's generative AI chat answered poorly on documents full of charts and tables. Uploading files by hand took about 15 minutes per task.
Approach
Dai-Dan, an engineering firm for air conditioning, electrical and IT infrastructure, ran a generative AI chat on Azure OpenAI with retrieval over internal regulations and technical documents. It added SoftBank's TASUKI Annotation tool to structure documents in-house. The tool also automated downloading, processing and uploading files.
Results
Takeaway
Dai-Dan's accuracy problem came from document structure, and a data preparation tool addressed it.
Categories
- Industry
- Construction and engineering
- Business function
- Internal work and knowledge management
- Innovation type
- Process innovation
- AI technology
- Generative AI
- Adoption
- Off-the-shelf tool or platform
- Evidence
- Vendor case study
Sources
Source summaries
SoftBank's customer story on Dai-Dan's internal generative AI system. It reports accuracy rising to 80% on figure-heavy documents and the removal of a 15-minute manual upload step.
Full details are in the original sources.