Agentic AI in Healthcare [in person]

This month we are going to have an in person meeting about AI in Healthcare. BI and Fabric, and also a review of the most important updates recently on the Power BI and Microsoft Fabric, and the AI-related topics about it. Join us for this session of learning and networking.
Agenda:
5:00-5:30 PM Welcome and Networking/Refreshments
5:30-6:30 PM AI in Healthcare
6:30-7:00 PM Wrap Up
Topic 1: Agentic AI in Healthcare: The RAG-Powered Diagnostic Assistant
\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\
Parametric LLM memory cannot support clinical deployment: no provenance, no patient-specific grounding at inference time, and unacceptable hallucination rates. Retrieval-Augmented Generation solves this by grounding an agentic system's reasoning loop in dynamically retrieved, evidence-backed context instead of static model weights.

The architecture: Chunking and embedding heterogeneous clinical data (imaging reports, discharge summaries, structured EHR fields), dense vector indexing with approximate nearest-neighbour retrieval, and context injection into the perceive-reason-act-learn cycle.

Key trade-offs: Retrieval latency versus context window budget, chunk size versus semantic coherence, embedding model selection for domain terminology, and multi-agent coordination for ranking and routing evidence across specialist sub-agents.

Results: 94.1% diagnostic accuracy in liver cancer detection; door-to-balloon time reduced from 211 to 68 minutes in a real cardiac workflow via concurrent retrieval across imaging, history, and evidence.

Open problems: Explainability at the retrieval-to-reasoning boundary, bias propagation through retrieved context, and the validation gap between proof-of-concept and clinical-grade reliability.
\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\

About the speaker: Dr. Saranya Selvarangan
A seasoned academic professional consultant, and data science professional with over 10 years of experience in teaching, research, and consultancy across Information Syste