- Funding Programme
- Year
- 2024
Advanced Programme in Natural Language Processing and Generative AI
The European Commission strengthened the Banco de España’s capacity to use modern Artificial Intelligence (AI) and Natural Language Processing (NLP) tools in supervisory work. Through a combination of expert training and the development of a Proof of Concept (PoC), the initiative enhanced the institution’s ability to analyse textual information efficiently, extract relevant insights, and support supervisory decision-making.
The project also provides a model that can be replicated by other National Competent Authorities (NCAs) facing similar challenges in processing complex financial information.
Context
Financial supervisors routinely analyse extensive volumes of textual information, including audit reports, supervisory documents and institution-level disclosures. As these datasets grow in complexity, supervisors face increasing challenges in identifying key information quickly and consistently.
Recent advances in Large Language Models (LLMs) and NLP techniques offer new possibilities for improving document analysis, semantic search and summarisation. For supervisors, these technologies can reduce time spent on repetitive manual reading while maintaining high standards of quality and oversight.
The Banco de España sought to strengthen its internal skills in these areas and explore practical use cases through hands-on experience and the development of a functional proof of concept.
Support Delivered
The support delivered consisted of two main components:
1. Expert Training Programme
A comprehensive 16-day training programme covered:
- Traditional NLP
- Deep learning and transformer models
- Fine-tuning and evaluating modern AI models
- Retrieval-Augmented Generation (RAG) and LLM applications
The training combined lectures, demonstrations, and practical coding exercises tailored to supervisory use cases. Participants from BdE gained practical, up-to-date skills to apply in real projects.
In addition to the technical training, the programme included a dedicated senior management strategy workshop focused on a concrete supervisory use case, aimed at discussing strategic and organisational implications of AI adoption.
Beyond technical upskilling, the training also aimed to foster an innovation-oriented mindset within the institution, supporting informed and responsible adoption of emerging AI tools.
2. Proof of Concept (PoC)
The PoC demonstrated how AI can assist in answering the annual external audit questionnaire using the audit report as the only source of information.
The PoC used semantic search and LLM-based generation to:
- Retrieve relevant passages from the audit report
- Produce draft answers to specific question types
- Support yes/no questions, extraction tasks and summary requests
- Provide traceability between answers and source text
The PoC did not use confidential supervisory data and served as a safe, replicable experiment for future development.
Result achieved
The project delivered the following key results:
- Improved technical capacity: BdE experts developed practical skills in NLP, LLMs and applied AI workflows.
- Operational insights: The PoC demonstrated that AI can meaningfully support tasks such as document triage and initial drafting of responses.
- Efficiency gains: AI-assisted retrieval and summarisation reduced the time required to locate relevant information in audit reports.
- Replicability: The methodology and findings can be adopted by other NCAs with similar supervisory processes.
- Foundations for future work: The project provides a basis for potential pilots and further exploration of AI in supervisory tasks.
More about the project
You can read the documents related to the project here: