The European Regulatory Copilot for Real-Time Compliance Monitoring

Organisations involved

Main Participant: Kodex AI is aGerman SME developing AI solutions for the European financial sector. Founded in 2021, it builds domain-specific language models that automate compliance, reporting, and risk management for financial institutions. By combining high-performance computing and generative AI, Kodex AI delivers precise, scalable tools that simplify complex workflows across evolving regulatory frameworks.

The challenge

Financial institutions must navigate an ever-expanding and constantly evolving framework of European regulations. Compliance teams face the overwhelming task of monitoring thousands of legal documents, interpreting highly specialised terminology, and producing accurate reports under tight deadlines. Existing commercial tools often provide fragmented or generic insights and struggle with the multilingual, technical, and context-dependent nature of regulatory texts.

This inefficiency increases operational costs, slows organisational responses to regulatory change, and raises the risk of compliance failures. Kodex AI identified that general-purpose language models such as GPT-4 cannot reliably interpret domain-specific regulatory language or provide the transparency required for audit scrutiny. To close this gap, the company developed the European Regulatory Compliance Copilot, powered by a fine-tuned Regulatory Large Language Model (RegLLM). Delivering such a system required identifying an appropriate base model, assembling large multilingual datasets, and conducting extensive experiments under strict accuracy and explainability constraints.

Kodex AI found that training and validating models of this scale exceeded the capacity and cost envelope of commercial cloud providers. The large-scale distributed computation, high-throughput data processing, and repeated iteration cycles required were only achievable through access to HPC resources.

The Solution

Kodex AI developed the RegLLM as a customised variant of leading open-source models designed specifically for European financial regulations. Using EuroHPC infrastructure MareNostrum5, the team trained RegLLM on more than 35,000 regulatory documents, guidance notes, and historic compliance reports. A continuous-ingestion pipeline keeps the model’s knowledge base up to date as new directives and regulatory updates are published.

A human-in-the-loop fine-tuning workflow ensures legal precision, explainability, and verifiable results. The European Regulatory Compliance Copilot automates horizon scanning, policy mapping, and structured report generation, delivering accurate and transparent insights that financial institutions can rely on.

Impact 

The FFplus innovation study enabled Kodex AI to move from limited experimentation in the Cloud to full-scale RegLLM development on HPC infrastructure. This shift allowed the company to train a domain-specific model that outperforms both open-source and proprietary systems (including GPT-4) on finance sector-specific regulatory benchmarks. Clients now benefit from substantial operational efficiencies: compliance teams report 60–70% time savings and annual cost reductions of €1–1.5 million for mid-sized institutions. Automated processing of regulatory updates also reduces non-compliance risk and accelerates adaptation of regulatory changes.

Strategically, the project expanded Kodex AI’s target market by 150% and reinforced Europe’s digital sovereignty by delivering an open, transparent product developed entirely on EuroHPC infrastructure. The company has projected its revenue will double over the next three years and thus position itself as a leading European provider of AI-driven regulatory compliance solutions.

Benefits

• 80% reduction in manual data processing through a real-time ingestion pipeline spanning 10+ regulators.
• 60–70% time savings for compliance teams via automated reporting and obligation extraction.
• Model iteration cycles cut from weeks to days using 48,000 EuroHPC GPU hours.
• Lower hallucination rates, strengthening auditability and trust in AI-generated outputs.
• HPC-ready infrastructure and standardised data governance now embedded in Kodex AI’s operations.