JapanFinance and insurance2024 – ongoingAozora Bank
On-premises Aozora LLM trained on internal data for questions on internal rules
Aozora Bank and neoAI trained an on-premises LLM on internal data and reported 130% of the earlier accuracy
Problem
Aozora Bank wanted a model that understood its internal terms and clerical rules for corporate and retail banking. Cloud API models did not meet its data security requirements.
Approach
Aozora Bank and neoAI built an on-premises platform on an open-source LLM. They continued its pretraining on the bank's internal data. The model was provisionally named Aozora LLM. It was tested on an internal benchmark that asks questions about clerical rules and uses retrieval over bank documents.
Results
Takeaway
The result comes from a benchmark the two companies built themselves. The release names no production launch date.
Categories
- Size
- Large enterprise
- Industry
- Finance and insurance
- Business function
- Internal work and knowledge managementProduction and quality
- Innovation type
- Process innovation
- AI technology
- Generative AI
- Adoption
- Partner co-development
- Evidence
- Company's own claim
Sources
Source summaries
A republished press release on the Aozora Bank and neoAI on-premises LLM. It reports 130% of the previous accuracy on an internal benchmark and describes gains in recognizing in-house terms and loan product structure. Next steps are further accuracy work and use in corporate banking.
Full details are in the original sources.