Introducing MedCortex!

Release v1.0.0: MedCortex Public Demo

This is the first public demo release of MedCortex, an agentic AI analyst for medical research.**

This platform is architected as a complete, professional workspace designed to solve the "information synthesis headache." It moves beyond simple RAG by implementing a sophisticated, multi-step reasoning engine, inspired by the "Objective-driven" and "Phase"-based workflows seen in professional research plans (like the provided "Team Alignment Map" and "Research Plan" templates ).  

Key Features in this Release

This demo showcases the full, end-to-end user journey, from analysis to a final, verified deliverable:

Agentic "Analysis Breakdown"

MedCortex is a multi-step agent. As seen in the UI, it features an "Analysis Breakdown" that executes a transparent "chain of thought." It performs "Query Analysis" and "Query Decomposition" to build a reasoning plan, enabling it to answer complex, multi-hop questions.  

Multi-Tool Routing (Text & Table)

The agent intelligently routes its sub-queries to the correct tool:
Hybrid Search Tool: For unstructured text, using a custom signal-based reranker (FAISS + BM25) and a domain-specific MEDTE embedding model for superior medical context.

TableRAG Tool: For structured data, allowing the agent to analyze and query data locked inside PDF tables.  

In-Line Claim Verification

Trust is the core feature. Every claim in a generated answer is fact-checked against its source documents using a Natural Language Inference (NLI) model. The UI provides immediate, transparent feedback:

"VERIFIED" tags for claims that are factually supported. (See Image 6: "VERIFIED" tag)

"REFUTED" tags for claims that are extrapolated or not found in the source text. (See Image 3: "EXTRAPOLATED" label)

"Synthesis Studio" for Deliverables

The end goal of a "Research Plan" is a "Deliverable." The "Synthesis Studio" (aka "Research Report") is the tool for this. Users can curate their verified findings from the chat into a single document and Export as DOCX or Markdown. This feature turns hours of manual report-writing into minutes of curation.

Enterprise-Grade Tech Stack

The platform is powered by IBM watsonx.ai and Granite models and is fully containerized for deployment on IBM Cloud Code Engine.

Try the Live Demo

You can test the full, end-to-end workflow—from the multi-step "Analysis Breakdown" (Image 4) to curating a final "Deliverable" in the "Synthesis Studio" (Image 7)—on our public demo.  

The application is deployed on IBM Cloud Code Engine.

Demo is available here: MedCortex