ChatBI is difficult to implement, let's see SwiftAgent empowering enterprises to analyze data in a new way
2024-08-07
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
At present, digital transformation has become a keyword for enterprise development, but many enterprises have encountered many difficulties in practical applications. The emergence of big models has become the key to changing the fate of enterprises, accelerating the arrival of data universalization and greatly lowering the threshold for data analysis. With the application of big models, many ChatBI products have been derived from the big model + BI approach. Currently, ChatBI on the market usually uses the NL2SQL technical path, that is, directly generating SQL through a large language model. This solution is prone to low data query accuracy (the accuracy is 60%-70%, and it will be lower if cross-table queries are made) and inconsistent data calibers.
Shushi Technology is an industry-leading data intelligence product provider with professional insights and technical strength in the fields of big finance, high-tech manufacturing and pan-retail, and has always been committed to using data and AI technology to solve corporate problems. The intelligent analysis assistant (SwiftAgent) launched by Shushi can effectively solve the difficult problems of ChatBI, making it possible for everyone to be a data analyst.
Taking the case of cooperation with Shushi Technology as an example, we can see the important role SwiftAgent plays in enterprise data analysis. During the digital transformation of a city commercial bank, it encountered problems such as the inability of existing data products to quickly produce in-depth conclusions end-to-end, complex demands for repeated reports, temporary demands becoming bottlenecks, and inconsistent indicator calibers. Based on the big data foundation, Shushi Technology completed the implementation of the indicator platform, unified all indicator calibers and data source lineages, and built the SwiftAgent intelligent data analysis assistant to lower the threshold for employee data analysis.
In business analysis, we often ask questions like: "How is our bank's credit data recently?" This is a very vague question. Does "recently" mean the last week or the last month? In addition, banks have various industry plans and special terms. For example, SDR (Special Drawing Rights), special terms need further explanation. For example, the word "performance" has different definitions and explanations for different roles.
Shushi Technology's solution takes into account the time dimension, industry-specific terms, and even understands ambiguous semantics, predicts the questions that employees may ask at different time periods, recommends common questions, and builds a personalized GASO model. Finally, it can also provide planning suggestions like a professional analyst to help employees complete the decision-making process more efficiently.
In summary, SwiftAgent by SwiftTech can solve many problems encountered by enterprises in the field of data analysis and decision-making, such as confusing data caliber, lack of data talent, high threshold for data use, long data analysis cycle, and inability to empower business decisions.
The new generation of SwiftAgent2.0 has also achieved five major highlights on the basis of the old version, including the construction of a unified semantic layer, establishing industry standards, indicators, people, goods and venue labels and other easy-to-understand semantic layers, solving the problem of unified data caliber in various departments of the enterprise, and effectively avoiding the phenomenon of dirty and messy data; users can intervene and guide users in a more natural way. Users can also ask questions again according to the prompts, and finally get the analysis content they really want to see. Users can also use the feedback of "likes" and "dislikes" for reinforcement learning, constantly correct errors, adjust queries, and make analysis more accurate; continuous reflection and learning can deposit all past question and answer analyses of users into the knowledge base, and directly provide conclusions and thinking processes in similar inquiry scenarios of other users in the future; multi-source data linking realizes multi-source heterogeneous data access, which can not only connect to data warehouses, but also import unstructured knowledge such as text, Excel, pictures, audio and video to meet comprehensive analysis ideas; data calculation acceleration engine, using the data calculation acceleration engine originally created by Shushi Technology, can achieve second-level data query and truly realize real-time human-computer interaction.
