CoolFace
Apppublic

Shahiniuz/Data_Explorer

sourceHugging Faceupdated 10mo agoView on Hugging Face
0likes
App README

CDC Diabetes Health Data Explorer

An AI-powered tool for querying real CDC health survey data using OpenAI's function calling API. Built for my capstone project.

About the Data

I'm using real CDC data from the 2014 Behavioral Risk Factor Surveillance System (BRFSS). It's a huge dataset - 253,680 actual patient survey responses with 21 different health indicators. You can verify this is real public health data from the UCI repository: https://archive.ics.uci.edu/dataset/891

The dataset includes things like:

  • —Diabetes status (binary - yes/no)
  • —BMI, blood pressure, cholesterol levels
  • —Lifestyle factors (smoking, physical activity, diet)
  • —Demographics (age, sex, education, income)
  • —Other health conditions

How it Works

The key idea here is privacy - the AI never sees the actual patient data. Here's the flow:

  1. 1.User asks a question like "what's the diabetes rate among smokers?"
  2. 2.AI gets the database schema (just table/column names)
  3. 3.AI uses function calling to decide which tools to use
  4. 4.Tools execute locally on your machine
  5. 5.Results go straight to the user

The AI is basically just writing SQL queries based on the schema, then the queries run locally. Patient data never leaves your computer.

Function Calling Implementation

This project uses OpenAI's function calling API with 3 tools:

Tool 1: execute_sql_query

  • —Executes SELECT queries on the database
  • —Has safety checks to block DELETE, DROP, etc.
  • —Returns results as JSON

Tool 2: get_database_statistics

  • —Gets overview stats (total patients, diabetes rates, etc.)
  • —Useful for general questions about the dataset

Tool 3: create_support_ticket

  • —Creates a support ticket when users need help
  • —Logs the ticket to console with timestamp
  • —AI can suggest this if it can't answer something

The AI decides which tools to call based on the user's question. Sometimes it calls multiple tools in sequence.

Setup

Requirements

  • —Python 3.9+
  • —OpenAI API key
  • —Streamlit

Installation

bash
# Install dependencies
pip install -r requirements.txt

# Download the CDC data (this might take a minute)
python3 download_real_data.py

# Create .env file with your API key
echo "OPENAI_API_KEY=your_key_here" > .env

# Run it
streamlit run app.py

Then open http://localhost:8501

Database Schema

The main table is patient_health_data with 253,680 rows. Some key fields:

Binary indicators (0 or 1):

  • —Diabetes_binary - has diabetes/prediabetes
  • —HighBP - high blood pressure
  • —HighChol - high cholesterol
  • —Smoker - currently smokes
  • —PhysActivity - physically active
  • —Plus more...

Numeric fields:

  • —BMI - body mass index
  • —GenHlth - general health rating (1-5)
  • —Age - age group (1-13)
  • —MentHlth - days of poor mental health (0-30)
  • —PhysHlth - days of poor physical health (0-30)

There are also 2 views (pre-computed aggregations):

  • —diabetesbyage - diabetes stats by age group
  • —healthrisksummary - health metrics by general health rating

Safety Features

Important for a healthcare data project:

  • —Only SELECT queries allowed
  • —Dangerous operations (DELETE, DROP, ALTER, etc.) are blocked
  • —All queries logged to console
  • —Query validation before execution
  • —Patient data never sent to OpenAI

Example Questions

Try asking things like:

  • —"Show me diabetes rates by age group"
  • —"What's the relationship between BMI and diabetes?"
  • —"How many patients have high blood pressure?"
  • —"Compare health stats between smokers and non-smokers"
  • —"Get overall database statistics"

Screenshots

Will add screenshots here showing:

  • —Main interface
  • —Function calling in action
  • —Query results
  • —Support ticket creation

Console Logging

Everything is logged to the console so you can see what's happening:

==================================================
Starting CDC Diabetes Data Explorer
==================================================
2025-11-13 12:00:00 - INFO - User asked: What's the diabetes rate?
2025-11-13 12:00:00 - INFO - Schema retrieved
2025-11-13 12:00:00 - INFO - Calling OpenAI API...
2025-11-13 12:00:01 - INFO - Calling function: get_database_statistics
2025-11-13 12:00:01 - INFO - Stats retrieved - 253,680 total patients
2025-11-13 12:00:02 - INFO - Getting final response...
2025-11-13 12:00:03 - INFO - Done processing query

Project Structure

Capstone1/
├── app.py                    # Main application
├── download_real_data.py     # Script to download CDC data
├── diabetes_health.db        # SQLite database (10MB)
├── requirements.txt          # Python dependencies
├── .env                      # API key (you create this)
└── README.md                 # This file

Tech Stack

  • —Python 3.9
  • —Streamlit for the UI
  • —OpenAI GPT-4o-mini with function calling
  • —SQLite database
  • —Real CDC data from UCI ML Repository

Data Source

This uses real public health data from the CDC's Behavioral Risk Factor Surveillance System. The dataset is available through the UCI Machine Learning Repository and is in the public domain.

Citation:

  • —CDC BRFSS 2014
  • —253,680 survey responses
  • —21 health indicators
  • —https://archive.ics.uci.edu/dataset/891

Notes

  • —The AI only sees database structure, not actual data
  • —All queries run locally
  • —Function calling is properly implemented (not just text generation)
  • —Real healthcare data (not generated)
  • —Built for educational/capstone project purposes

Feel free to explore the code - it's all commented and should be pretty straightforward.