ThinkDeep’s Sovereign AI Agents Help Automate Public Services for the Government of France

DeepBrain allows domain-specific AI agents that are able to connect and interact with huge amounts of data from different sources in order to extract and analyze information to find relevant patterns. The DeepBrain agent’s unique data confidentiality approach guarantees the data stays on the customer infrastructure at all times, to ensure the highest level of security and confidentiality.

Through DeepBrain, users can have their own fraud detection agent or an agent for legal document processing, using domain knowledge aggregated from all the French legal documentation. As a multi-agent system, these agents can interact with one another to realize complex workflows or answer cross-domain queries.

ThinkDeep ensures peak performance by utilizing advanced NVIDIA technologies, including NIM for integrating, packaging, and deploying the newest large language models (LLMs) on premise, and NVIDIA Dynamo to optimize inference speed for state-of-the-art reasoning models. ThinkDeep also used the NeMo framework to seamlessly integrate LLMs and vision language models (VLMs) into its existing infrastructure.

To ensure the safe and responsible use of these powerful AI agents, ThinkDeep implemented robust guardrails using the NVIDIA NeMo Guardrails toolkit. The guardrails provide essential safeguards against potential risks such as biased or harmful outputs, helping them to maintain the highest standards of ethical AI practice.

To ensure data sovereignty and confidentiality, the Ministry of Economy and Finance selected NVIDIA for a sovereign on-premises infrastructure solution. The NVIDIA DGX™ platform allows the ministry access to the GPUs required to serve millions of daily users with the latest models.

The benefits of DeepBrain for the French government include faster document processing, cost savings, productivity gains, and energy savings.

• Customers were able to automate the document retrieval process, improving from two days to two minutes.
• Implementation was low effort, with just one week to set up DeepBrain instance and configure connectors.
• Productivity increased with 1,000 hours saved per month for a typical 100-user deployment.
• The Ministry of Finance has saved €2 million by deploying DeepBrain at scale for 10,000 employees.
• By controlling the models and infrastructure, the ministry optimized energy management and lowered energy management and environmental costs.

DeepBrain has significantly accelerated legal document analysis for public servants. What previously took up to two days to find relevant information now takes two minutes, allowing faster response times and saving millions of euros, both in operational costs and by detecting potential fraud faster and earlier in the process.

The DeepBrain Graph methodology enables real-time, effortless searching across millions of files and relationships, delivering sourced and precise answers. Simultaneously, it constructs custom, domain-specific knowledge bases adapted to individual use cases and users.

Since the launch of its DeepBrain product, ThinkDeep has experienced significant market traction and customer interest, particularly within the sovereign AI domain. The increased demand spans a wide range of industries, including finance, public sector organizations, aerospace, and energy. The sovereign AI capabilities of DeepBrain likely appeal to these sectors due to their need for data security, control, and compliance with local regulations.