SiaGPT: Generation of Service Agreements from Contracts

Our client provides a range of services to its own clients: 

• Setting up renewable heating and cooling networks that supply energy to cities, neighborhoods and industrial complexes, enabling them to heat and cool their buildings without emitting CO2. 
• Maintaining installations that produce heating, high-pressure steam, superheated water, compressed air, etc. while reducing resource use and carbon emissions over the long term. 
• Performing energy construction projects on industrial sites and buildings to renovate or install new heating equipment (boilers, heating network pipes, plant equipment, etc.) and building components (insulation for walls, roofs and windows, optimized ventilation, upgraded materials, etc.). 

Internally, our client models all these services through service agreements objects (CVS in French) in its IT system, which provides a summary of the various contractual commitments it owes to customers. 

Contract 

• The contract, drawn up by the client, defines the contractual commitments sold to its customers (municipality, private client, etc.) 
• The services provided in the contracts are of several types: services sold, performance criteria (contractual temperatures), reasons for solicitation by the customer.   

Service agreement 

• The service agreement (CVS) is a structuring of contractual information. 
• It makes it possible to describe the contractual services in summary form.  
• The service agreement lists the geographical locations in which the various services apply. 

After a new contract has been signed and characterized by the relevant business department, the business integrator receives the contract for validation. During this validation, the business integrator reads the contract and the appendices to model several objects in the tools of the repository. This modeling was done manually, without assistance and required the business integrator to read and process the whole contract. 

The use of artificial intelligence was expected to save a considerable amount of time in the modeling of service agreements and therefore more generally in the overall modeling of the objects in the repository. 

With our experience in generative AI and semantic search, we developed a solution to automate the retrieval of information from our client’s contracts to establish a service agreement (CVS), based on semantic search through documents, combined with the use of our SiaGPT tool to extract information within documents. 

This solution consists of the implementation of a user interface where the user can upload files representing a contract, to generate the associated CVS Excel file. The solution was successful in automating extraction for all file types (scan or pdf contract) and contract types. It also demonstrated that new generative AI models far outperform traditional NLP models while reducing implementation time.