Challenge

Securex, a leader in human resources and social administration, provides a broad range of services across multiple domains, including Securex's SEPP (Service Externe de Prévention et de Protection) department. SEPP offers critical workplace safety and compliance advisory services, relying on a vast archive of documents for regulatory compliance, health and safety guidelines, and expert recommendation reports.

Despite having a structured document repository in SharePoint, Securex faces significant challenges in retrieving relevant historical documents when writing advisory documents. Employees often relied on informal networks such as WhatsApp to ask colleagues for similar documents, leading to inefficiencies. Furthermore, an extensive backlog of unstructured advice remained uncategorised, making it nearly impossible to systematically search for and reuse prior insights.

The manual process of document tagging and summarisation was labour-intensive and would take an estimated two and a half years to complete if done manually. Securex needed an automated solution to streamline document classification, improve retrieval efficiency, and enrich metadata with relevant summaries and keywords.

Solution

Faktion has developed an AI-powered document management system for Securex' AI Lab, integrating advanced Natural Language Processing (NLP) techniques for:

  • Document classification: AI-based tagging to categorise documents within Securex’ structured SharePoint repository, based on the in-house classification scheme.
  • Summarisation: automated generation of concise summaries to provide quick insights into document contents.
  • Dynamic keyword extraction: identifying key terms to enhance metadata and improve searchability.

By leveraging Large Language Models (LLMs), the system effectively classified documents with high precision. The solution also provided intelligent summaries tailored to Securex' domain expertise and extracted critical keywords, allowing employees to quickly identify and retrieve necessary documents.

Key features of the system include:

  • Automated metadata enrichment to enhance document search and retrieval.
  • Integration with SharePoint to ensure seamless access within Securex' existing infrastructure.
  • User feedback loop enabling Securex domain experts to validate and refine AI-generated outputs, ensuring continuous improvement.

Approach

To ensure the effectiveness of the AI-powered solution, Faktion followed a structured methodology:

  1. Data analysis & preprocessing: extracting and preprocessing textual data from Securex' SharePoint archives. Subsequently, we applied Argilla as an annotation tool for quality validation and domain expert feedback
  2. Model development: for the classification and summarisation, we tested several LLMs and applied few-shot and zero-shot techniques. Afterwards we compared different models on classification/summarisation accuracy and computational efficiency. For the dynamic keyword extraction, we used YAKE and evaluated its performance against LLM-generated keywords
  3. Evaluation & optimisation: we conducted human-in-the-loop validation by Securex experts to ensure high relevance, clarity, and groundedness of AI generated summaries and extracted keywords. We also measured performance metrics such as precision, recall, and F1-score for the classification.
  4. Implementation & integration: we integrated the AI system into Securex' SharePoint environment for real-time document categorisation and retrieval. Subsequently, we developed an interactive UI allowing users to review and refine AI-generated tags and summaries

Outcome

The AI-powered document management system delivered significant efficiency gains and improved knowledge accessibility at Securex. Some key numbers:

  • 80%+ accuracy in document classification, exceeding initial benchmarks
  • 50% reduction in document tagging and summarisation time, significantly improving operational efficiency
  • 100% trust rating from Securex evaluators for the groundedness of AI-generated summaries
  • Enhanced document retrieval and searchability within Securex' SharePoint repository.

By replacing manual document classification with AI automation, Securex can now rapidly access relevant knowledge, optimise its regulatory compliance workflows, and streamline its advisory services. The system also lays the foundation for future AI-driven knowledge management, positioning Securex at the forefront of digital transformation in workplace safety and compliance.