Case Study: Technical Office Chatbot

This initiative originated from one of our Automation Audits conducted at [redacted], a company generating €9 million in revenue in the thermo-hydraulic plant manufacturing sector. The analysis revealed a need to reduce the amount of time the technical office spent on repetitive, low-value checks. The main objective was to eliminate time waste, accelerate feasibility assessments, and provide reliable answers to both internal and external requests.

The chatbot comes into play during the technical analysis and design phases, where — based on input from the technical office — it verifies feasibility against production specifications and workshop capabilities (machinery, technical drawings, etc.). It is also useful in after-sales, as it allows users to quickly query the system to obtain information about the sold plant and any technical limitations. The primary users are technicians and sales staff.

The system’s knowledge base is built on technical manuals, operational documentation, and historical records of non-conformities. These sources are updated every six months, and access is limited to the internal company network, restricted by IP-based constraints.

Automation Description

The system adopts a Retrieval-Augmented Generation (RAG) approach, allowing the AI to process a much larger volume of data than a simple chatbot fed with PDFs. Starting from the technician’s specifications, the model retrieves the most relevant content, analyzes it, and provides a feasibility assessment accompanied by reasoning, practical alternatives, and potential risks. Operational guidelines prevent the model from executing instructions beyond its functional scope, ensuring both consistency and safety.

Currently, the process involves submitting the technical sheet associated with a drawing and verifying compliance with workshop specifications. An extension is planned to enable automatic drawing analysis for physical composition description, followed by a feasibility evaluation on the generated summary. CAD drawings are textually described by the AI and indexed within a vector database, making them searchable and reusable for future responses.

The interface is browser-based and accessible within the local network, usable from both handheld devices and PCs. This setup ensures broader accessibility for both workshop and office staff including non-technical personnel without depending on external connections or third-party tools.

The system is deployed on-premise: data never leaves the client’s infrastructure and is not sent to third parties. Access is restricted to the corporate network, guaranteeing control and confidentiality throughout the entire usage cycle.

Results

The adoption of this automation significantly reduced verification times: what previously required three days within a 22-day prototyping cycle has been reduced to a single two-hour meeting between production and the technical office. The saved time has been reallocated to higher-value activities, improving both lead time and decision quality through clearer reasoning, alternatives, and explicit risk analysis. Human intervention remains essential for final decisions and defining alternative solutions, maintaining technical control without sacrificing AI-driven efficiency.

The next step is the implementation of an automatic drawing analysis pipeline to generate structured descriptions. This will allow the technical office to feed increasingly valuable data into the system while enabling the AI to process them as well-formatted text. In parallel, monitoring metrics will be introduced covering accuracy, latency, and automatic response rate to ensure ongoing control of quality and performance over time.