Dr.Guido

Elisabetta Borghi é representante da Dr.Guido, uma plataforma que utiliza inteligência artificial para oferecer assistência médica virtual inicial de forma rápida, eficaz e permanente. A empresa tem trabalhado para garantir a precisão dos diagnósticos e a personalização da assistência médica, por um lado, e a proteção da privacidade dos utilizadores, por outro. Nesta entrevista, realizada durante a Web Summit, Elisabetta Borghi explica detalhadamente como funciona esta app e em que fase está a empresa. #InteligênciaArtificial #WebSummit #Saúde #DrGuido

Using AI Agents in medical use cases can be tricky Here's how we can approach it better... Though AI Agents are very powerful for enterprise and personal use cases, Integrating agents or even LLMs in medical use cases has been very difficult One of the core reasons is that medical data is highly complex, multifaceted, and often unstructured. Solution? A hierarchical multi-agent architecture that integrates LLMs with automated knowledge graph construction for medical diagnosis. 📌 Let me explain the architecture: 1. Knowledge Graph Construction Pipeline - Transforms raw medical documents into a structured graph of medical entities (diseases, symptoms, treatments) and their relationships. - This process consists of several sub-steps: a. Data Chunking & Segmentation: Splits medical documents into smaller chunks for better processing. b. Entity & Relationship Extraction: Uses BioBERT (biomedical LLM) to extract entities and relationships, mapping them to medical standards like SNOMED-CT and UMLS. c. Graph Construction: Builds a graph with entities as nodes and relationships as edges. d. LLM Augmentation: General-purpose LLMs enhance the graph by identifying additional context-aware relationships. e. Human Validation: Experts review and expand the graph to ensure accuracy and completeness. 2. Hierarchical Multi-Agent Framework: - Simulates real-world medical diagnostics where general practitioners collaborate with specialists for comprehensive care. - This step also consists of sub-steps: a. GP-LLM (General Practitioner Model): Acts as the first contact, analyzing queries and providing initial diagnoses. b. Consultant-LLMs: Specialized models trained in fields like cardiology, neurology, or endocrinology handle domain-specific cases. c. Referral Mechanism: Refers cases to specialists if GP-LLM confidence is low or expertise is needed. d. Inter-Agent Communication: Ensures smooth information exchange and collaborative diagnosis refinement. The architecture makes it easier for healthcare professionals to use Agents to diagnose the possible disease. 📌 A few of the key features of this architecture include: 1. Mirroring Real-world Medical Workflows: The hierarchical agent structure simulates the collaboration between general practitioners and specialists, allowing for efficient handling of complex cases and targeted use of expertise. 2. Hallucination Mitigation: The multi-agent verification and knowledge graph constraints help reduce the risk of LLMs generating incorrect or misleading information. This is a rather fascinating way to look at how we can improve health care using AI Agents The architecture is from an amazing paper by Kaiwen Zuo, Yirui Jiang, Fan Mo and Pietro Lio Check the comments to learn more What are your thoughts about this architecture? Let me know in the comments below 👇 Please make sure to, ♻️ Share 👍 React 💭 Comment To help more people learn