SPIE - Automating Ad-hoc questions
Open Learning
Semester programme:Open Learning/Innovation
Client company:SPIE
Jip Voss
Joy van Rijn
Margarita Fedulova
Yasmin Abhrao Kfuri
Guilherme Pinto de Oliveira
Project description
Our client for this project is company SPIE Industry Services specializes in engineering, construction, maintenance, and optimization of industrial processes and systems. The Business Control department aims to improve efficiency and effectiveness through AI-driven solutions.
With the main question: How can an AI chatbot enhance the role of the business controller at SPIE by improving the efficiency and speed of data analysis and reporting?
Problem statement: the volume of available data is increasing, while fewer people are available to analyze and interpret it. This creates a growing need for automation and intelligent AI-driven solutions. The Business Control department wants to spend less time on routine tasks and enhance the speed and accuracy of data analysis.
Context
This software engineering project is set within SPIE Industry Services, specifically their Business Control department, which seeks to enhance data-driven decision-making using artificial intelligence. As the volume and complexity of data continue to grow, business controllers are under increasing pressure to deliver faster and more accurate analyses. The project responds to this challenge by developing an AI-powered chatbot aimed at streamlining data interpretation, report generation, and ad-hoc analysis.
The objective is to build a functional proof of concept (PoC) that integrates seamlessly with SPIE’s internal data systems and assists controllers by automating routine tasks. Key technologies include Natural Language Processing (NLP), data integration pipelines, and Retrieval-Augmented Generation (RAG), all tailored to the financial and operational context of SPIE.
A notable feature of the proposed solution is a layered AI architecture involving multiple agents that iteratively analyze data for deeper insights, correlations, and anomalies going beyond static report generation. This innovative, explorative approach aims to proactively support business decisions, suggesting additional data collection and continuously learning from user feedback.
Security, compliance, and scalability are essential considerations, ensuring that the solution meets industry standards and can evolve with future needs. The success of this project will be measured by the chatbot’s usability, data accuracy, and the degree to which it empowers business controllers to shift focus from routine reporting to strategic analysis.
This context sets the foundation for an impactful AI solution that modernizes business control within a complex industrial environment.
Results
The developed solution is an AI-powered chatbot system designed to support business controllers at SPIE by automating data analysis and reporting tasks. The product leverages Python for the back-end—ideal for data processing and AI integration and Streamlit for the front-end, offering a user-friendly web interface.
At this stage, the prototype demonstrates a fully integrated workflow between the user interface and the data processing back-end. The user interacts with the system by asking natural language questions through the front-end. These queries are sent to the back-end, where the input is processed and translated into a DAX (Data Analysis Expressions) query. This DAX query is then injected into SPIE's data warehouse to retrieve the relevant structured data.
Once the data is retrieved, it is passed to a Large Language Model (LLM), which interprets the results in the context of the original question. The model generates a clear, human-readable response with contextual analysis. This final output is then delivered back to the front-end, allowing the user to receive insights in real-time.
The system currently supports basic end-to-end interaction with actual data, demonstrating that the concept works in a realistic environment. The integration of core components NLP processing, DAX query generation, data retrieval, and analytical response has been successfully tested with real-world datasets, marking this prototype at Technology Readiness Level 6: a system/subsystem model or prototype demonstration in a relevant environment.
This result proves technical feasibility and sets the stage for further refinement, including improved model fine-tuning, security implementation, and scalability planning.