AgentiCDS: A Multimodal Agentic AI System for In-Hospital Clinical Decision Support | Farah Shamout
Title history
- AgentiCDS: A Multimodal Agentic AI System for In-Hospital Clinical Decision Support | Farah Shamout — observed on 20261104-2221 UTC (YouTube)
Keywords
Summary
124 words
Critical Evaluation
Value of the Information & Strength of the Argument
The talk provides valuable insights into the challenges of deploying agentic AI in clinical settings, backed by experimental results. The argumentation is coherent, with clear comparisons and evidence from benchmarks on MIMIC and the AgentRX dataset. The speaker also discusses limitations and open questions, such as reliance on single-agent performance and the rapid evolution of models, demonstrating scientific rigor and balanced reasoning.
Scientific Rigor, Source Quality, Title Accuracy
The scientific rigor is high; the speaker is an academic and references known datasets like MIMIC, as well as under-review papers and public benchmarks. The only direct source cited is the ADIA Lab Symposium link, but the talk mentions multiple research outputs. The title matches the content perfectly, as the entire presentation revolves around the AgentiCDS system and its components.
137 words
Title / Content Match
The title accurately reflects the content, as the talk focuses on a multimodal agentic AI system for clinical decision support.
Quality & Reliability
8/10
The presentation is based on rigorous academic research, referencing public datasets like MIMIC and submitted papers, but external sources are limited to the conference link and claims are not independently verified.
Key Moments
Markers derived by PSI from the transcript: the creator did not define chapters.
- Introduction of speaker and talk overview
- Presentation of project objectives and system architecture
- Discussion of long-context reasoning challenge and experiments on MIMIC
- Introduction of AgentRX benchmark and multi-agent reasoning
- Analysis of failure modes in multi-agent systems
- Future work: multi-site evaluation and safety layer
- Q&A session on human-in-the-loop and agent improvements
Cited Sources
- ADIA Lab Symposium 2026 — Conference where this talk was presented, providing context for the session on Health Sciences.
Contribution & Novelties
The talk contributes original research in applying agentic AI to clinical prediction tasks, specifically introducing the AgentRX benchmark and analyzing failure modes in multi-agent systems compared to single-agent approaches. The novelty lies in rigorous evaluation of long-context and multi-agent reasoning for ICU data, and the identification of architecture-dependent performance based on patient subgroups. The work also highlights the practical challenges of integrating multimodal data in clinical environments and suggests future directions for safety and reliability.
Pour aller plus loin :
- Multi-agent system — Relevant to the multi-agent reasoning challenges discussed.
- Electronic health record — Context for the EHR data used in the system.
- MIMIC III database — The public ICU dataset used in their experiments.
115 words
Radar Profile
The radar profile shows a high niveau_technique score (9) and high fiabilite_globale (8), with moderate quantite_information (8) and qualite_information (8). This indicates a technically dense and reliable presentation, though the quantity and quality are consistent without excess, suggesting a focused and credible overview.