Capitalize on a patrimony of 100 million documents with Generative AI

What was the need?

Artelia is an engineering and consulting group specializing in construction, infrastructure, water, industry and the environment. The Group has 8,900 employees and sales of €934 million in 2022. Operating internationally, the company relies on technical expertise, project management and innovation to meet the complex challenges of its customers. In this context, Artelia’s interest in generative artificial intelligence (AI) is part of an optimization and innovation approach, in particular for the management of its vast document library.

Faced with the increasing digitalization and complexity of engineering projects, ARTELIA is faced with a major challenge: the management and operation of a colossal document patrimony, amounting to over 100 million documents. This immense library includes a variety of formats and data types, from technical drawings and study reports to correspondence and administrative documents, reflecting the richness and complexity of the projects carried out by the group.

The Generative AI solution provided

In the space of 5 months, JEMS and SNOWFLAKE developed a natural language conversational interface (LLM) that enables the user to dynamically query the corpus of documents and obtain results in real time. The chatbot relies on precision algorithms that have been specifically trained to understand and process queries related to these unstructured documents.

The practical application of this technology has resulted in two main use cases:

• Instant retrieval of documentary evidence, particularly in the legal field, where fast and accurate access to specific information can be crucial. Thanks to generative AI, users can retrieve users can find relevant documents without having to manually navigate through millions of files.
• Automatic generation of new content for calls for tender, using models trained on historical data, such as the Cahiers des Clauses Techniques Particulières
  (CCTP) of previous projects. This not only saves time in the creation of new documents, but also ensures consistency and quality, based on well-established
  precedents.

These solutions have significantly improved the efficiency and quality of ARTELIA’s business processes, by operating the power of generative AI to transform a massive, unstructured body of documentation into a living, easily accessible and usable resource.

Technologies used

• Snowflake: A cloud data platform that facilitates the management, sharing and secure analysis of large-scale data.
• Streamlit: A tool for creating data applications in Python quickly and with little code.

Project duration

• 8 months with 3 FTEs

Missions carried out

• Production of a Proof of Value (POV) for the chatbot based on generative AI, to concretely demonstrate the benefits of this solution prior to wider deployment.
• The implementation of a data-centric architecture, which promotes the integration, quality and availability of data for optimal use in decision-making processes.
• The implementation of two use cases, which validated the effectiveness of the approach and the usefulness of the chatbot in ARTELIA’s day-to-day operations.