SQL Data Extraction and Analysis with a Multi-Agent AI Solution for Industrial Printing Machines

SQL Data Extraction and Analysis with a Multi-Agent AI Solution for Industrial Printing Machines

Introduction: The Challenge of Data Accessibility in Industrial Printing

Like in many industries, industrial printer manufacturers have decisive competitive advantage: their knowledge about what their customers do and how they do it. Their crux is: This knowledge is hidden in their vast telemetry databases and one would need data experts to know what data were available and how to interpret them. What the partners wanted was a system that would help users from the clarification of what they actually needed all the way to providing an answer to the user’s initial question.

Solution Overview: AI-Powered Data Retrieval

Both parties used Erium’s approach of Cognitive Process Modelling to co-develop: Extensive experimentation and refinement, for which the Halerium AI platform proved to be an invaluable enabler, ensured the AI’s reliability and accuracy and significantly improved the level of assistance it can provide.

Erium’s solution involves an AI assistant that allows not only engineers but also business experts to access data from all connected printers in a cloud data lake without manually writing SQL queries. Users can simply ask questions, the AI system first delievers the most relevant headlines from to the request from the database, then generates the necessary SQL queries and visualizations. This innovative approach ensures that even those without IT expertise can retrieve, analyze data efficiently and return insights burried in the database.

Towards a Team of AI Agents:

Problem: Overloaded 1-Bot approach

With its extensive experience in genAI, Erium recognizes that a classical 1-bot approach will fail to guarantee the accuracy and precision required for user interactions with such a massive data.

Solution: Divide & Conquer Approach

For an efficient and robust solution, Erium implemented a multi-agent system. This system comprises specialized agents, each with distinct roles:

  • Requirements Engineer: Understands user input and translates it into actionable tasks.
  • SQL Expert: Generates SQL queries based on user requirements.
  • Business Intelligence Engineer: Visualizes data, creating plots and tables.

This approach distributed responsibilities among various agents, dramatically improving efficiency and reducing errors. Each agent’s specialization meant quicker, more accurate responses tailored to specific tasks.

The Balance of Capability and Cost Efficiency

Problem: “Intelligence” Requires Costly Compute

The next step was to optimize the system’s performance and cost of querying the entire SQL schema for each user query.

Solution: Vector Store Technology Implementation

Erium implemented vector store technology to streamline the process by conducting semantic searches within the SQL schema. The requirements engineer agent now receives the most relevant section of the SQL schema rather than the entire dataset. This adjustment slashed token consumption by 96%, significantly cutting costs and speeding up response times.

Additionally, the SQL Expert agent was provided with query pattern cases to expedite and standardize query creation, ensuring that queries met user needs efficiently.

The Result: A Seamless, Scalable Solution

Through continuous iteration and refinement, Erium delivered a comprehensive, multi-agent chatbot system. The final solution not only improves data extraction speed and accuracy but also integrates smoothly with the customer’s existing infrastructure. By ensuring the system’s scalability and maintainability, Erium laid the foundation for future enhancements and expansions.

Remarkable Time Savings and Eye-opening Insights

At a first glance, the AI solution has resulted in a significantly higher usage of the database and 60% faster time “time to information”. But beyond that, users commented that it has helped them, when “we did not even know that we had this data”.

Conclusion: Transforming Data Management

Besides all the hype about AI technology, the things that were fundamental to the success of this project could not be farther from the technology itself: complete focus on the user and the involvement of the business experts to infuse their know-how into the AI from the very start.

Organizations should look to partner with technology providers like Erium and carefully choose those with a project approach focused on leveraging their knowledge. If you would like to expand on the above and find out more, don’t hesitate to get in contact with Erium. Discover how Halerium enables you to build AI solutions yourself that can streamline your operations, save time and enhance your productivity. Contact us today at Halerium.

Project Tasks

  • Chatbot for SQL Query Generation

    We developed and integrated an intelligent chatbot that interprets and clarifies user requests, automatically generates SQL queries to retrieve the correct data and visualizes it. For this we had the business experts infuse their know-how about the data and technology to the greatest extent possible, giving the AI the best possible preparation to help the user.

  • Transparency for system maintenance and risk management

    We allow multiple AI agents to focus on specific aspects of the user question and collaboratively work out the answer for the user. This greatly reduces the complexity of each agent and enables business experts to maintain and audit the entire system.

  • Vector Store Technology for Resource Optimization

    We implemented a vector store architecture that feeds the right pieces of information to each agent based on a semantic search. This has cut running cost of the system by more than 95%.

Project Overview

An industrial printing company established quick and easy access to insights in their unique machine data base for business users. The multi-agent AI solution that was developed in the Halerium AI platform, was codeveloped with the customer’s business experts leveraging Erium’s approach of Cognitive Process Modelling. The multi-agent AI system facilitates access to data-based insights, which will greatly enhance decision making across business segments.

Outcomes
  • Multi-agent system for Database crawling, SQL querying and Data vizualisation for all business experts
  • Information management architecture reduces costs by 96%, optimizing resource use
  • Cooperation between AI and business experts for faster, better technological solutions.
  • LLM Knowledge Transfer to Customer's R&D Team