AI-Powered chatbot
Driving Digital Transformation with Generative AI by Leveraging ChatGPT and Azure OpenAI to Transform Product Documentation Queries.
Customer
Bystronic
Industry
Machinery Manufacturing
Services
Development, Support
Technologies
ChatGPT within Azure Open AI services, ReactJs, LangChain, Azure Cognitive Search
About Bystronic
Bystronic is a global leader in the manufacturing industry, specializing in the production of bending and laser-cutting machines. Operating in over 30 countries with production and development centers in Europe, Asia, and the US, it employs over 3,500 people. As part of its digital transformation journey, one initiative focused on enhancing customer support and service through innovative AI technology solutions.
The Challenge
Tech Info Scattered Across Files Causes Delays
Bystronic’s product information was spread across various formats, such as websites, Excel sheets, Word documents, and PDFs. This made it difficult for customers and support teams to quickly find the needed information, causing delays and frustration. Each format had unique challenges—for example, some were difficult to search, and others were difficult to organize correctly.
The team faced three key problems:
- Bringing together all the different formats into one easy-to-access system.
- Managing the highly detailed and technical nature of product information.
- Making sure people could quickly find the correct information, even if it were buried deep in complex documents.
The Solution
An AI Chatbot Using Azure OpenAI and ChatGPT to Handle Customer Queries
Bystronic implemented a proof of concept (PoC) using Azure OpenAI Services, ChatGPT, to develop a chatbot capable of retrieving information from the company’s technical product documentation.
Our Approach
Custom Data Ingestion Methods
Different document formats, such as web content, Excel sheets, Word documents, and PDFs, posed data extraction and indexing challenges. PDFs often had complex layouts requiring careful parsing, while Excel files contained structured tabular data that needed to be accurately represented. We developed custom data ingestion pipelines tailored to each format to manage this, normalizing the data into a consistent schema. We then used a vector-based database (azure cognitive search) to enable semantic indexing, allowing for high-precision, context-aware querying across all datasets. This approach ensured the chatbot could retrieve technical product information efficiently, regardless of document type or format, significantly improving accuracy and response relevance.
Optimizing AI for Technical Queries
With highly specific and technical product descriptions, prompt engineering enabled accurate responses. We continuously refined the chatbot’s output by optimizing prompts and leveraging natural language processing (NLP) techniques. By combining Azure Cognitive Search with ChatGPT, we built a system to interpret complex product queries and improve user conversation flows.
Utilizing LangChain Frameworks
Large language models (LLMs) like ChatGPT can sometimes generate responses that lack context or stray from the intended scope. We implemented prompt engineering techniques and structured workflows to address this and guide the model toward more predictable and relevant outputs. By incorporating the LangChain framework, we created a robust pipeline that manages the model’s interaction with the client’s technical data. LangChain enabled us to structure complex multistep processes, ensuring the chatbot delivered consistent, context-aware responses while interacting with various data sources. This approach improved the chatbot's performance and enhanced the user experience by maintaining conversation context and accuracy over time.
Ensuring Data Security with a Secure GPT Deployment in Azure OpenAI
When dealing with sensitive technical documentation, Bystronic highly values data security. We configured a secure GPT deployment within Azure OpenAI to meet these requirements, leveraging Azure's enterprise-grade security features. This setup included data encryption, network isolation, and role-based access controls, allowing Bystronic to control data access and comply with strict privacy regulations. The solution ensured that all product data queries and interactions remained private and secure while maintaining the high performance and efficiency of the chatbot. This configuration provided the necessary safeguards for managing proprietary information.
The Result
Improved Customer Support Efficiency
The PoC demonstrated that customers could quickly retrieve detailed product information by interacting with the chatbot, reducing manual search times. This would free human agents to handle more complex inquiries, showing how the system could streamline support processes.
Higher Customer Engagement and Satisfaction
In tests, the chatbot showed the ability to provide fast, accurate responses to customer inquiries, suggesting that it could enhance customer experience and reduce the workload for support teams once fully implemented. This positions the chatbot as a promising extension of Bystronic’s knowledge base.
Increased Business Intelligence
The PoC provided valuable insights into using AI models to query technical documentation, enabling Bystronic to discover additional optimization areas like data ingestion and prompt engineering.
Further Outlook
An AI-Powered Digital Metaverse Integration
This project demonstrated the significant impact of using conversational AI and ChatGPT for querying complex technical product documentation. The collaboration between our team and Bystronic has optimized customer support and set the stage for future AI-driven solutions that will further enhance customer engagement.
Following the success of the PoC, Bystronic is now working on integrating the chatbot into its Digital Metaverse Initiative using Microsoft Mesh. This futuristic solution allows customers to engage with product documentation in a fully immersive 3D environment, taking their customer interactions to a new level and providing a cutting-edge experience far beyond rule-based chatbots.