Introduction
Reporting has always been a key part of answering business questions. Dashboards and analytical reports are a great way to delve into data analysis and draw conclusions and recommendations for decision-making. However, they require a level of understanding on the data to interpret these results and the presentation can get confusing and crowded.
Project Overview
ProCogia’s innovation team has taken the initiative to find a solution for this problem by taking advantage of Large Language Models.
With the right database architecture, it is possible to train an AI agent to translate a business question from plain English to SQL and provide the data required back to the user.
To maintain data security, the AI agent does not have access to the data itself, instead, it acts as a translator between human and database.
After solving the translation problem, a simple process to send the query to the reporting database should suffice to provide the user a tabular or visual report they need.
Key Factors for Success
LLMs are great at parsing text and retrieving information from data but are not capable of thinking for themselves. For this reason, the reporting database needs to be ready for the module to use, the concept of garbage in and garbage out also applies in this product.
How can we prepare a reporting database to provide reliable data for the business?
- Understand your business problem and what questions need to be answered. This can be achieved by a series of meetings with key departments of the organization and their experts.
- Engage with the technical team to provide the main business questions.
- The technical team will use these questions to craft a schema for the reporting database.
- After the reporting database objects are in place, data will need to be cleaned and loaded into its respective objects, ETL jobs will be scheduled with its respective business rules in place.
At this point, the reporting database is ready to be connected to the AI agent. The technical team can create and provide the document with the reporting database schema parameters, the business questions (with semantic variances), and the translation into SQL.
Test Cases and Data Retrieval Success Rates on Documented Data
If a question has been documented in the module and the correct database schema, and requirements definition have been set, the innovation team has achieved over 80% success rate on retrieving the correct data for questions that the module has been trained on.
Technology
These are the technologies used as prototype, but the tool can be integrated in any environment and cloud provider.
Front End:
Backend:
Ready to Empower Your Teams with QueryIQ?
Don’t let data bottlenecks hold your organization back. Discover how QueryIQ can transform your team’s access to insights and drive impactful results.
👉 Schedule a Demo Today and see QueryIQ in action!
