Use Case Retail
Exploring Internal Data
This project has the goal of providing a general and adaptable solution. In the following section, a specific use case is presented, which was developed as part of a hackathon.
Given the nature of the hackathon, a functional demo was created to showcase the concept. However, it has been indicated that this project is expected to be further developed in the near future, according to the established timeline.
Challenges
Explore and use internal data of the company that is difficult to use with regular data analytics tools.
Create a powerful LLM
To construct queries to the client’s DBs
Auto generated queries
Run these auto generated queries and retrieve data from the DBs.
Comprehensive information
Process such response and provide comprehensive information that answers the user made query.
NRT
The solution must be in NRT.
Solution
Use any of the available LLMs to tackle these challenges. Important to note that the solution must be in Near Real Time (NRT), so some LLMs do not provide fast answers. We made use of any posible model such as GPT3.5, GPT4, Claude, Gemini-pro and other state of the art solutions.
Use a vectorstore database to include similar results that the LLM should output based on the user made questions. These provide a significant improvement in the output of the LLM. Finding relevant input & expected output is crucial.
We developed a “chatbot” so you can ask in natural language to the system.
Tech stack
Results
The data now is being used. Before it was not. We do not have any KPIs. But there is a major improvement in the time that the users need to find data.
