nicole

Developed project for automating the generation of credit rating reports, reflecting the risks of bonds for over 10,000 listed companies in early stages. Simplified project by creating streamlined categorization. Utilized expert analysis to assess financial data across industries, delivered over 15,000 detailed reports within 1.5 hours. Designed non-standardized data extraction tools, utilized multi-modal models to extract data from complex border-less tables in audit reports and prospectuses, incorporating both text and image data. Developed diverse datasets for multiple table formats, utilized GNNs for encoding, and predicted text relationships with post-processing for table reconstruction. The generating quality of simple table headers is high, with an accuracy rate of 97%, while the accuracy rate for complex table headers reaches 94%. Developed a text extraction module of regulatory inquiries based on Transformer model with an accuracy rate of 95%.

Developed a 3D floor plan service, optimized deep learning algorithms, and integrated augmented reality technology. This pioneering project received high acclaim in the architectural field and was awarded the Ping An Emerging Technologies Award. Designed and implemented distributed data processing services and machine learning/deep learning models leveraging Spark, encompassing text classification, text summarizing and advanced modules for further enhancing analytical capabilities.