The context, the need 📍

The integration of generative artificial intelligence into industrial processes marks a strategic step forward for companies seeking to gain in efficiency and innovation. As part of the Generative AI Program, Alstom, world leader in intelligent and sustainable mobility, has collaborated with ILLUIN Technology to deploy three generative AI solutions, adapted to the complex challenges of the rail industry.

The rail industry relies on rigorous, often complex processes, involving large volumes of technical documentation, rapid and precise responses to calls for tender, and high standards of standardization and quality.

 

The approach and key success factors 🏆

ILLUIN Technology implemented tailor-made solutions, integrating Large Language Models (LLM) and advanced document analysis tools. Each project was based on a methodology centered on :

• • • Close collaboration with Alstom's business teams
    • MVPs (Minimum Viable Products) to rapidly validate concepts and refine functionalities
    • Particular attention paid to the user experience through intuitive interfaces

The 3 completed projects ✅

 

1. Pseudo-code generation in specification documents

 

The need 💬

Alstom engineers have to draw up numerous software design specification documents for each new train. In these documents, they must manually maintain consistency between various pieces of information:

• • specifications written in natural language
  • pseudo-code (using a specific internal standard)
  • definition of variables and parameters

This work is time-consuming and error-prone.

 

The solution 💡

ILLUIN Technology has created an intuitive interface with :

• • A pseudo-code generation engine: to generate pseudo-code directly from natural language specifications and the list of available variables.
  • A requirements search engine: to enable the engineer to write a draft specification and search the database for nearby existing specifications and pseudo-codes.

The results 🙌

• • Reduce specification writing time
  • Improved consistency between specifications and pseudo-code tests

 

Technical stack 🤖
  .     

 

2. Automated extraction of technical information

 

The need 💬

Alstom's engineers spend a lot of time consulting technical documentation in order to draft new documents.
The aim of the project is to provide an MVP capable of automating, accelerating and facilitating information extraction via :

• • • A simple user interface
    • The option of asking a single question or a list of questions
    • Relevant visualization of results, downloadable in Excel format

The solution 💡

ILLUIN Technology has implemented a sourced synthesis pipeline capable of answering various types of questions from a corpus of complex technical documents parsed using the ILLUIN Document Pipeline including tables, images, flowcharts, etc.
The solution can also automatically populate an input file with hundreds of questions.

 

The results 🙌

Solution currently in production for a group of testers, to be expanded over time.

 

Technical stack 🤖
  .      .

 

3. Generation of tender responses

 

The need 💬

Alstom's business teams responsible for responding to requests for proposals (RFPs) have to process large volumes of PDF documents. The process of responding to a call for tenders is often tedious and time-consuming, which explains the desire to mobilize generative AI to automate certain tasks. Business teams have identified several stages in responding to a call for tenders where AI could bring value:

• • • summary of AO documents
    • structured extraction (csv format) of technical requirements; expressed in the AO and classified according to a business nomenclature
    • matching between the requirements extracted in the previous step and what has been done in Alstom projects;
    • pre-generation of slides fed by the generated summary, matching what's required with what's been done at Alstom on previous rail projects.

The solution 💡

For this first exploration, the focus was on the first two steps outlined above. For the first item, we set up a pipeline to summarize tenders in a structured way according to predefined axes. For the second, we explored the potential of vision models for extracting text from complex PDF documents (with figures, colors, graphics). 

 

The results 🙌

Potential validated with the business, the project is intended to be industrialized.

 

Technical stack 🤖
.

 

Conclusion

The partnership between ALSTOM and ILLUIN Technology reflects a shared ambition: to integrate cutting-edge artificial intelligence solutions to transform complex challenges into strategic opportunities. Thanks to this collaboration, ALSTOM confirms its status as a visionary leader in the rail industry, ready to adopt disruptive technologies to optimize its processes and strengthen its competitiveness.

 

Useful links 🔗

🦹 Our bespoke tech achievements in AI, GenAI and Data Science
🚀 Watch Alstom's conference / feedback on large-scale GenAI deployments