—
Total Queries
—
Avg Confidence
—
Avg Latency
—
Answer Found Rate
Recent Queries
Clinical RAG Assistant
MIMIC-IV · ClinicalBERT · FAISS · LLaMA 3 · Research Use OnlyBio_ClinicalBERT
Loading...
⚠️ For experimental/research purposes only — de-identified MIMIC-IV data — not for clinical decision making
Cardiology
Clinical
Neurology
Genomics
Pharmacy
Ask a clinical research question
Powered by 130M+ de-identified MIMIC-IV notes · Research use only
Press Ctrl+Enter to send
0 / 2000
Analytics Dashboard
Query performance and usage statisticsSystem Evaluation
Retrieval quality and performance metricsConfidence Metrics
Mean Confidence—
Min Confidence—
Max Confidence—
High Confidence Rate (>75%)—
Latency Metrics
Mean Total—
Mean Retrieval—
P95 Total—
Answer Quality
Found Rate—
Not Found Rate—
Total Evaluated—
System Info
Vectors Indexed—
Embedding Model—
LLM Model—
Device—
About
System information and disclaimer⚠️ Important Disclaimer
This system is for experimental and research purposes only. It uses de-identified MIMIC-IV clinical notes from PhysioNet. This tool is NOT intended for clinical decision making.
🔬 What This System Does
ClinicalNote AI is a Retrieval-Augmented Generation (RAG) system that searches 130M+ de-identified MIMIC-IV clinical notes to answer research questions using semantic search and a large language model.
🏗️ Architecture
Data130M+ MIMIC-IV clinical notes via Apache Spark
EmbeddingsBio_ClinicalBERT (768-dim, GPU)
Vector SearchFAISS — 951,725 indexed vectors
LLMLLaMA 3 via Ollama (local)
APIFastAPI async + PostgreSQL logging
GPUNVIDIA RTX 5060 Ti — 17.1GB VRAM
👨💻 Developer — Sohel Ahmed, PhD
LinkedIn
linkedin.com/in/sohelcu06
GitHub
github.com/sohel10
Email
sohelcu06@gmail.com
Live Demo 1
energyforecastai.org ↗
Live Demo 2
ClinicalNote AI (this app) ↗
📊 Data Source
MIMIC-IV (Medical Information Mart for Intensive Care) is a publicly available de-identified healthcare database. Access requires PhysioNet credentialing.