Medical NER Dashboard

A Gradio web application that extracts medical named entities from 1,000 clinical notes, maps them to standard medical codes, renders an interactive dashboard, and provides a chatbot to query the results — all running locally with no external API keys required.

Dataset: ncbi/Open-Patients

Clinical Bot

Select number of records to Analyze

NER Dashboard

Once the data is loaded, you can start chatting with the bot. Download the summary

What it does

Step	Description
1	Loads 1,000 records from ncbi/Open-Patients (HuggingFace)
2	Extracts conditions, symptoms, medications, and procedures via keyword NER
3	Looks up the top 10 of each category against free NLM APIs: ICD-10-CM, RxNorm, CPT, and SNOMED-CT
4	Renders an interactive Plotly dashboard (bar charts + code tables, dark theme)
5	Provides a chatbot (Ollama) to ask questions about the extracted data

---

Requirements

• Python 3.10+
• Ollama running locally

Supported models (select in the Chat tab dropdown)

Display name	Ollama ID	Pull command
Llama 3.2	llama3.2	ollama pull llama3.2
GPT OSS 20B	gpt-oss:20b	ollama pull gpt-oss:20b
Gemma 3 1B	gemma3:1b	ollama pull gemma3:1b

Pull at least one model before starting:

ollama pull llama3.2   # recommended default (~2 GB)

Python dependencies

gradio>=4.0.0
ollama>=0.1.7
pandas>=2.0.0
plotly>=5.18.0
datasets>=2.14.0
requests
tqdm

Install all at once:

pip install -r requirements.txt

---

Quick start

# 1. Clone and enter the repo
git clone <repo-url>
cd chatbot_healthcare

# 2. Create and activate a virtual environment
python -m venv .venv
source .venv/bin/activate          # Windows: .venv\Scripts\activate

# 3. Install dependencies
pip install -r requirements.txt

# 4. Make sure Ollama is running with llama3.2
ollama serve          # or open the Ollama desktop app
ollama pull llama3.2

# 5. Launch the app
./launch.sh           # or: python gradio_simple.py

Open http://localhost:7860 in your browser.

---

Usage

Tabs

Tab	What to do
Data Analysis	Click "Load & Analyze Data" to run NER and see a text summary
Chat with Bot	Ask questions about the data (e.g. "What are the most common conditions?")
Dataset	Check available records; clear the NER cache to force a re-run
NER Dashboard	Adjust the record slider, click "Run NER Dashboard" for the Plotly chart
Export	Download results as CSV files

Using a different Ollama model

OLLAMA_MODEL=mistral ./launch.sh

How to run

Clean port 7860 and start Gradio

lsof -ti:7860 | xargs kill -9 2>/dev/null python gradio_simple.py

Run the dashboard

python Medicalnerdashboard.py

NER cache

Results are cached at cache/ner_hf_biomedical.pkl after the first run (~1 min for 1,000 records). Subsequent runs load instantly. Use the Dataset tab or delete the file to force a fresh run.

---

Architecture

gradio_simple.py          <- Gradio UI (all tabs, event wiring)
medical_ner_dashboard.py  <- Core pipeline library
  load_texts()            <- HuggingFace datasets loader
  run_ner()               <- d4data/biomedical-ner-all transformer NER
  lookup_all_codes()      <- NLM ICD-10-CM / RxNorm / CPT / SNOMED APIs
  build_dataframes()      <- pandas DataFrames per category
  build_figure()          <- Plotly 4x2 subplot dashboard
  _build_context()        <- text summary fed to chatbot
launch.sh                 <- startup script (checks Ollama, kills port 7860)
requirements.txt          <- all Python dependencies

---

Medical code sources (all free, no API key)

Code system	Used for	API
ICD-10-CM	Conditions, symptoms	NLM Clinical Tables
RxNorm	Medications	NIH RxNav
CPT	Procedures (curated map)	Built-in lookup table
SNOMED-CT	Procedures (fallback)	NLM Clinical Tables

---

Key files

File	Purpose
medical_ner_dashboard.py	Core NER + code-lookup pipeline
gradio_simple.py	Full Gradio web UI
launch.sh	One-command launcher
requirements.txt	Reproducible environment

---

Submission checklist

• Loads 1,000 records from ncbi/Open-Patients
• Extracts conditions, symptoms, medications, procedures
• Top 10 per category with ICD-10-CM / RxNorm / CPT / SNOMED-CT codes
• Interactive dashboard (Plotly bar charts + code tables)
• Chatbot grounded on extracted data (Ollama, no API key needed)
• Fully local — no external API keys required