Knowledge Chat for Veolia Energia Polska

________________

Case Study Summary

Veolia Energia Polska engaged WEBSENSA to address the challenge of improving the accessibility and retrieval of information from their company documents. Initially focused on technical documents within the tech department, the scope of the project quickly expanded to other areas, including HR and internal knowledge distribution.

WEBSENSA developed the Knowledge Chat – an innovative AI-powered chatbot that revolutionised Veolia's internal knowledge management. Leveraging advanced technologies such as Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG), this solution enabled employees to quickly obtain both technical as well as HR company information.

As a result, Veolia experienced streamlined operations, enhanced decision-making, and reduced costs. Interested? Continue reading to gain more in-depth insight into the Knowledge Chat application.

________________

GenAI-Powered Knowledge Chat for Veolia

Leveraging Large Language Models to Enhance Information Retrieval from Company Documents

WEBSENSA partnered with Veolia Energia Polska – a leader in resource management, environmental protection and public services, to develop an innovative conversational chatbot solution based on their internal document base.

Veolia Faced Challenges in Streamlining Information Access

Veolia's primary challenge was navigating numerous company documents that were difficult to search and access. This issue first emerged in the tech department, where specialists struggled to access technical documentation. We then identified the opportunity to extend the project to include other documents related to HR processes and regulations.

Our approach to solving this challenge was to create a specialised chatbot that provides Veolia employees with precise answers to technical queries (such as troubleshooting, safety protocols, and maintenance procedures) as well as HR-related processes (such as delegation handling, insurance, and benefits). The goal was to streamline access to crucial information and enhance Veolia's operational efficiency.

Veolia's challenge was employees' difficulties with access to relevant answers about internal company processes

We Developed an AI Chatbot for Efficient Knowledge Retrieval

The solution we created to address Veolia’s challenge was the Knowledge Chat. It is aimed to enhance operational efficiency by navigating complex regulations and procedures in the company’s documents and providing the employees with essential organisational knowledge.

The Knowledge Chat transforms company documents into an easily searchable, interactive knowledge base. Thanks to the solution, users can ask questions directly through the chat interface and receive precise answers based on the content of the uploaded documents.

The system always indicates the source document of the information, ensuring transparency and facilitating further analysis.

Agile Methodology Enabled Seamless Project Execution

We followed an agile project management methodology, ensuring clear and transparent communication through regular in-person and virtual meetings.

Using Jira and a scrum methodology with 2-week sprints, we executed the project through 4 key stages: identifying Veolia’s needs, configuring an LLM architecture, developing a responsive chat interface, and rigorous testing and fine-tuning.

The team had open discussions about issues and cooperated with the client to solve them." — Maciej Dzięcielak, Architect IT, Veolia Energia Polska

We Used LLMs, RAG and NLP for Robust Knowledge Management

For the chatbot to function effectively, we created an advanced AI-driven environment capable of supporting the deployment of Large Language Models (LLMs) and a Chat GUI for Retrieval-Augmented Generation (RAG).

Key deliverables:

• LLM Architecture Preparation and Configuration – establishing a robust and scalable architecture to support advanced language models,
• User Interface Development – creating a conversational chat interface with context-based search capabilities using natural language processing (NLP),
• Context-Based Search Implementation – employing NLP for efficient organisational knowledge management.

The technological stack:

• Python (Django, Langchain) – for backend development, ensuring a scalable server-side application,
• PostgreSQL – as the primary relational database for reliable data storage,
• Qdrant – for vector search, enhancing the chatbot’s search capabilities,
• Large Language Models (LLM) – core to AI functionality, enabling natural language understanding and response generation,
• Vue – for the frontend web framework, facilitating a responsive chat interface,
• Google Cloud Platform (GCP) – hosting the entire application, ensuring scalability, security, and high availability.

Our Key to Success Was Avoiding Common LLM Pitfalls

The innovative approach to the Knowledge Chat implementation involved integrating advanced features to ensure a robust, secure, and adaptable solution.

1. Preventing AI Hallucinations – We implemented measures to minimise incorrect or irrelevant information, ensuring precise and accurate responses.‍
2. Grounding – Knowledge Chat uses only Veolia's data, ensuring that all responses are specific and applicable to employee queries.‍
3. Granular Permissions – We defined user access levels to data and knowledge, so only authorised users can access sensitive information.‍
4. Jailbreak Prevention – We established high-security standards and measures to prevent jailbreak attempts, ensuring Veolia’s data security.‍
5. Multiple and Dynamic Data Sources – Knowledge Chat connects seamlessly to various data sources, updating information in real time, ensuring the most relevant answers.‍
6. Adaptable Behaviour – The system tailors responses to different contexts and user needs by adjusting settings, prompts, and configurations, enabling its use in different scenarios.

The Solution Improved Veolia's Efficiency and Decision-Making

The collaboration significantly improved Veolia’s digital capabilities:

1. Enhanced Knowledge Management – Knowledge Chat facilitated efficient knowledge search in Veolia, ensuring quick access to the information needed.
2. Increased Organisational Initiatives – The chatbot enabled Veolia’s employees to develop new business features based on knowledge from more accessible documents.
3. Operational Efficiency – Knowledge Chat reduced the time employees spent searching for information and allowed them to focus on more crucial tasks.
4. Better Decision-Making – With easy access to accurate information, Veolia’s managers can make more informed and timely decisions.
5. Reduced Costs – Knowledge Chat helped reduce costs associated with knowledge management, such as training, document storage and IT support.

In the first 2 months post-deployment, the chatbot successfully handled approximately 2500 queries from around 400 users. A wider rollout is expected to provide more comprehensive statistics, further demonstrating the solution's value and impact.

Client Appreciated Our Transparency and Engagement

Veolia's review of our work on Clutch:

• Quality: 5.0
• Schedule: 4.5
• Cost: 5.0
• Willing to Refer: 4.5

"WEBSENSA completed all tasks with full transparency and high engagement." — Maciej Dzięcielak, Architect IT, Veolia Energia Polska

We Continue Collaborating to Expand the Solution Further

WEBSENSA continues collaborating with Veolia on the Knowledge Chat, integrating additional knowledge sources and enhancing functionalities such as document inference for procurement support. These improvements aim to further optimise the solution’s quality and operational costs.

Contact Us

Are you interested in implementing a similar solution for your business? Contact us – our experts are ready to tailor our offer to your specific needs.