LegML
LegML develops compact, sector-specialized large language models for regulated industries such as finance and law. Its flagship Hugo 32B model for French law claims 10% higher accuracy than a frontier model while using half the parameters and 75% less compute. The company also offers Le Lion, a free demo platform, and Mo', a voice AI assistant for Microsoft Teams.
- Artificial Intelligence
- Legal Technology
- Software Only
- AI Agents
- Enterprise Software
- Financial Technology
Funding
Founders
Founder details are not available yet.
Product
Problem
General-purpose large language models often underperform on specialized tasks in regulated industries such as finance and law, where domain-specific terminology, compliance requirements, and accuracy are critical. Organizations in these sectors struggle to find models that deliver reliable results without the cost and complexity of deploying much larger, resource-intensive systems.
Solution
LegML develops compact, sector-specialized LLMs engineered to outperform generalist models up to five times their size on industry-specific benchmarks. Each model is trained on rigorous domain benchmarks, such as French business law and Dutch legal standards, ensuring relevance and precision. The company provides a free demo platform, Le Lion, for users to test models on their own use cases, with enterprise deployment support available for production integration. The training methodology separates sector adaptation, which uses public and synthetic material, from client-specific deployment, where models operate within the customer's environment without touching their data. Additionally, LegML offers Mo', a voice AI assistant for Teams that provides automatic transcription and summaries, extending AI capabilities into everyday workplace workflows.
Target Audience
Primary customers are enterprises in regulated industries, including legal firms, financial institutions, and compliance-heavy organizations, as well as teams using Microsoft Teams who need AI-powered meeting assistance.
Features
- Sector-specific LLMs for finance, law, and regulated industries, with dedicated models like Hugo 32B (French law) and Le Lion Affaires (French business law), curated and evaluated to outperform generalist models on domain benchmarks
- Compact model architecture that balances performance and efficiency: Hugo 32B claims 10% higher accuracy than a frontier model while using half the parameters and 75% less compute, trained in 14 days
- Free demo platform (Le Lion) for testing models on real business use cases, with an insurance-law assistant launching early November 2026 that cites clauses and generates claims letters
- Custom benchmark development based on client needs for new sectors or specialized requirements
- GDPR compliance with data processing in Europe, ensuring digital sovereignty for regulated industries
- Mo' voice AI assistant for Teams with automatic transcription and meeting summaries
- Open-source contributions on Hugging Face to advance AI in regulated sectors