🩺 EquiCare AI: Diagnostic De-Biasing Engine

🔴 View Live Demo Here (Note: Replace with your actual Streamlit Community Cloud link)

Securing equitable healthcare through intelligent, real-time algorithmic auditing.

EquiCare AI operates as a robust algorithmic governance middleware designed to detect, audit, and mitigate demographic bias in predictive healthcare models. It forces transparency by intercepting black-box diagnostic algorithms and utilizing Large Language Model (LLM) orchestration to ensure medical decisions are equitable and free from historical data skews.

---

🚀 See It In Action

The AI Governance Engine Under the Hood

Our Streamlit dashboard seamlessly transitioning between proactive data auditing and reactive patient diagnostics.

---

🛑 The Problem: "Invisible Algorithmic Harm"

Predictive medical AI models often carry fatal demographic biases inherited from historical training data, leading to a massive crisis in clinical equity.

• 📉 Demographic Skew: Datasets are heavily weighted toward specific demographics (e.g., relying on male physiological data for cardiac events).
• 👻 Atypical Erasure: Black-box algorithms frequently misdiagnose or ignore atypical symptoms in underrepresented groups.
• 🔇 Algorithmic Gaslighting: Patients receive inaccurate "Low Risk" scores, leading to wrongful discharges and severe clinical harm.

💡 The Solution: Dual-Phase AI Governance

EquiCare AI replaces opaque algorithmic guessing with an Automated Governance Loop:

1. Pre-Model Audit 📊: Hospital admins input proposed training dataset parameters. The AI flags inherent demographic skews before training begins.
2. Clinical Interception 🩺: A physician inputs live patient data. If the hospital's black-box model outputs a suspicious score, the middleware intercepts it.
3. AI Trace 🧠: Google Gemini orchestrates a retrospective explainability trace, analyzing the clinical context against the flawed algorithm.
4. Correct & Notify 📲: The system generates a human-readable fairness report, explaining the algorithmic failure and providing a bias-corrected medical recommendation.

---

✨ Key Features

• 🔍 Proactive Data Auditing: Detects feature bias risks and underrepresented cohorts in raw historical datasets.
• 🛡️ Reactive Explainability Tracing: Real-time interception of black-box diagnostic outputs for instantaneous bias auditing.
• 🔄 LLM Orchestration: Powered by Google Gemini Flash-Lite to handle complex JSON data structuring and semantic medical reasoning with extreme low-latency.
• 📋 Automated Markdown Reporting: Generates structured, professional Fairness Audit Reports directly for attending physicians.

---

📸 System Previews

1. Phase 1: Pre-Model Data Auditing

Administrators input historical clinical dataset parameters for pre-training evaluation.

The LLM orchestrates a deep-dive analysis, flagging intersectional blindness and outputting actionable data-rebalancing techniques.

2. Phase 2: Post-Model Explainability Tracing

The middleware flags a potentially lethal "Low Risk" discharge order due to missing fairness metrics.

Gemini Flash-Lite generates a real-time audit, explaining the algorithmic failure and providing a bias-corrected clinical recommendation.

---

🛠️ Technology Stack

• Frontend & UI Rendering: Streamlit (Python)
• AI/Machine Learning: Google Gemini Flash-Lite API (via Google AI Studio)
• Data Structuring: Python Dictionaries, JSON
• Cloud Infrastructure: Streamlit Community Cloud

---

💻 Quick Start Guide (Local Deployment)

Prerequisites

• Python 3.9+
• Git
• Google AI Studio API Key

Installation Steps

1. Clone the repository:

   git clone https://github.com/symbiote07/equicare-ai-governance.git
   cd equicare-ai-governance

2. Install dependencies:

   pip install -r requirements.txt

3. Configure API Secrets: Create a hidden Streamlit secrets directory to secure your Gemini API key.

   mkdir .streamlit
   # On Windows use: type NUL > .streamlit\secrets.toml
   touch .streamlit/secrets.toml

   Open secrets.toml and add your key:

   GEMINI_API_KEY = "your_api_key_here"

4. Launch the Streamlit Dashboard:

   streamlit run app.py

   The dashboard will automatically open in your default web browser.

---

🤝 Open for Collaboration

This project is open-source and actively seeking contributors! Areas where we'd love your help:

• 🏥 EHR Integration: Building pipelines to integrate directly with hospital Electronic Health Records (EHR) via FHIR/HL7 protocols.
• 🧠 Specialized Model Swapping: Creating configuration files to easily swap from Gemini Flash-Lite to enterprise Vertex AI Pro models for production-grade clinical reasoning.
• 🌍 Multi-lingual Output: Utilizing LLM translation capabilities to generate audit reports for international medical teams.

Feel free to Fork the repository and submit a Pull Request!

---

Developed by Aryan Shukla (Team pennywise07) for robust, transparent, and equitable healthcare diagnostics.