Vontive is excited to share the results of a new Mortgage Underwriting LLM benchmark, produced in collaboration with Vals AI. The benchmark focuses on the performance of LLMs on document extraction and logical tasks: https://lnkd.in/gtZBbvfr ?? Key Highlights:????? Anthropic's Claude 3.7 Sonnet emerges as the top performer, achieving 80.6% accuracy and excelling in extracting semantic and numerical information from mortgage tax documents. ??? What makes this benchmark unique?????? a. Tested models' ability to process complex, multi-format documents. b. Evaluated both semantic extraction and numerical calculation tasks. c. Used a diverse dataset of 1,258 real-world mortgage tax certificates. ? ?? Evaluation Insights: ????? a. Models demonstrate the capability to understand both computer-generated and handwritten content. ????? b. Semantic extraction showed consistent performance, while numerical tasks revealed significant model variations. ????? c. Performance variations highlight the complexity of multimodal document processing. If you're interested in learning more about how Vontive and other leading FinTechs are leveraging LLMs in their workflows please join us in-person in Seattle or virtually on March 21st for Vontive's Innovation Forum! https://lnkd.in/ggfhqnvv
Vontive
软件开发
San Francisco，California 6,013 位关注者
Vontive is the first embedded mortgage platform for investment real estate.
关于我们
Vontive builds financial products for real estate investors, delivered through brands they trust. Our no-code, white-label solution empowers any business serving real estate investors to launch its own investment-property mortgage offering in weeks and deliver reliable debt capital to its customers.
- 网站
-
https://www.vontive.com
Vontive的外部链接
- 所属行业
- 软件开发
- 规模
- 51-200 人
- 总部
- San Francisco，California
- 类型
- 私人持股
- 创立
- 2017
地点
-
主要
获取路线
10 Jackson St
US，California，San Francisco，94111