• 🎯 Mission

    Design and develop an AI-powered application to help users of naval technical documentation quickly find and consult the technical requirements relevant to their operational context — by extracting content from large regulatory PDFs and enabling contextual, semantic search.

  • 🎖️ Result

    A performant web application backed by vector AI search, metadata extraction, and contextual filters — enabling users to retrieve, understand, and trace the right requirements in seconds instead of days or even weeks.

Context

As part of the VERGA project—jointly led by Thales and Beslogic and funded by Confiance IA—we supported the development of an AI-powered internal tool for professionals who regularly consult naval documentation to ensure regulatory compliance. These users must sift through vast collections of technical standards to validate applicability conditions and safety requirements on naval platforms.

Yet, navigating these lengthy, PDF-based documents was slow, manual, and prone to oversight. With increasing complexity and tighter project timelines, the need for a faster, smarter way to identify and consult relevant requirements became critical.

Challenge

How can we help users of naval technical documentation quickly find the right regulatory requirements within hundreds of pages of standards?

The tool needed to:

  • Support precise filtering (e.g. by applicability domains)
  • Enable natural language queries
  • Display contextual results without losing reference to the original PDF
  • Be fast, easy to update, and scalable for future document sets

Our Approach

 

Intelligent Data Extraction

Using custom pipelines built with tools like Docling and Unstructured, PDF documents are transformed into structured, queryable data. Each requirement is enriched with metadata like domain of applicability, chapter, and positional coordinates, then indexed in a vector database.

Hybrid Search Interface

We built a modern single-page web application (SPA) using Angular and AG-Grid to streamline requirement exploration. The interface offers two complementary search modes:

  • Structured filters by keyword, tag, or applicability domain
  • AI-powered search that understands natural language queries and surfaces semantically relevant requirements

The AI also assists in classifying requirements and suggesting relevant applicability domains based on user queries.

Technologies

  • Frontend: Angular, AG-Grid
  • Backend: Python
  • Processing: Docling, Unstructured
  • AI: LLMs, Elasticsearch, vector embeddings

Results

The VERGA platform enables users to:

  • Instantly locate relevant regulatory requirements using keyword filters or semantic AI search
  • Explore and compare extracted requirements, enriched with metadata and grouped by domain or relevance
  • Maintain full traceability to the original PDF documents, including page numbers and contextual placement
  • Reduce time spent on manual document review while increasing confidence in regulatory compliance

By combining natural language processing, metadata extraction, and contextual filters, VERGA transforms complex naval documentation into a responsive, intelligent consultation tool.

Collaboration & Trust

VERGA is a joint R&D initiative led by Beslogic and Thales, and funded by Confiance IA. Developed through close collaboration, the project brought together technical, scientific, and AI expertise from both teams. This partnership fostered mutual trust and laid the groundwork for a broader collaboration in software and AI innovation.

Back to Case Studies