Doctolib’s AI agents streamline healthcare support without sacrificing security

A healthcare technology company Doctolib has developed an AI-powered support system called Alfred that functions like a digital butler to handle routine customer inquiries while maintaining strict security protocols.

The big picture: Doctolib’s implementation of an agentic AI system represents a practical application of artificial intelligence in healthcare support, designed to process approximately 17,000 daily messages.

• The system, named Alfred, operates as a network of specialized AI agents working together, each with defined roles and specific tools at their disposal
• By handling routine support queries, Alfred enables human agents to focus on more complex cases requiring personal attention
• The system leverages LangGraph for orchestrating interactions between different AI agents and incorporates a RAG (Retrieval Augmented Generation) engine for improved response accuracy

Technical architecture: The implementation prioritizes security and reliability while maintaining efficient operation.

• Service-to-service authentication is managed through JSON Web Tokens (JWTs), ensuring secure communication between system components
• AI agents operate with inherited user permissions rather than administrative access, maintaining security boundaries
• The system includes fact-checking mechanisms and human-readable confirmations before executing sensitive actions
• Large Language Models (LLMs) are restricted from directly executing sensitive operations to prevent potential errors or security risks

Performance monitoring: Doctolib has established clear metrics to evaluate the system’s effectiveness and efficiency.

• Success is measured on a 1-3 scale to assess the level of achievement in handling support requests
• Efficiency metrics track both latency and the number of steps required to resolve inquiries
• The system’s ability to handle high message volumes (~17,000 daily) demonstrates its scalability

Real-world application: Alfred functions as an intelligent support system capable of understanding and addressing user needs effectively.

• The system can guide users through complex processes, such as managing calendar access rights
• User context is maintained throughout interactions to ensure appropriate and personalized responses
• The RAG engine enables accurate information retrieval and response generation based on existing knowledge

Looking ahead: While Alfred represents a significant advancement in AI-powered healthcare support, its current implementation serves as a foundation for future expansion.

• Doctolib continues to identify opportunities for expanding Alfred’s capabilities and use cases
• The successful deployment demonstrates the viability of agentic AI systems in handling sensitive healthcare-related support tasks
• Future developments will likely focus on enhancing the system’s ability to handle increasingly complex support scenarios while maintaining security and efficiency standards