Medical Intake Partner - System Overview

Description

An AI-powered medical intake assistant that conducts multilingual voice-based patient interviews, performs symptom triage, and generates clinical reports. The system uses Google's Gemini AI for intelligent questioning and data extraction, with support for voice interaction in multiple languages.

Key Features

Multilingual Support

• Languages: English, Hindi, French, Spanish
• Voice I/O: Speech recognition for input, text-to-speech for output
• Auto-translation: Seamless conversation in user's preferred language

Clinical Workflow

1. Patient Registration: Collects name, age, and gender with validation
2. Symptom Detection: AI identifies symptoms from natural language descriptions
3. Dynamic Triage: Adaptive questioning based on clinical relevance (red flags, severity)
4. Data Extraction: Synthesizes severity (1-10 scale), duration, urgency level, and clinical summary
5. Verification Loop: Allows corrections before finalizing report

Output & Storage

• PDF Report: Professional clinical intake document with patient profile and medical evaluation
• MongoDB Integration: Optional encrypted storage with patient data hashing
• Structured Data: JSON-formatted medical records for downstream systems

Technical Stack

AI & NLP

• Google Gemini API (gemini-3-pro-preview)
• Speech recognition (Google Speech API)
• Edge TTS for voice synthesis
• Deep Translator for multilingual support

Data Processing

• Async/await architecture for concurrent operations
• Robust JSON extraction with regex fallbacks
• Fuzzy matching for language selection

Storage & Export

• FPDF for clinical report generation
• MongoDB for persistent storage
• SHA-256 hashing for patient anonymization

Configuration

Required Environment Variables

GEMINI_API_KEY=<your_api_key>
MONGO_URI=<mongodb_connection_string>  # Optional
TEXT_MODE=0  # Set to 1 to disable voice

Dependencies

google-generativeai, edge-tts, speech_recognition
pygame, fpdf, deep-translator, pymongo, thefuzz

Usage Flow

Language Selection → Disclaimer/Consent → Bio Collection → 
Symptom Interview → AI Triage (per symptom) → 
Summary Verification → PDF Generation → Database Storage

Safety Features

• Medical disclaimer requirement before proceeding
• Confirmation loops for voice recognition accuracy
• Fallback to text input on recognition failure
• Urgency flagging (low/med/high) for clinical prioritization
• 30-second audio timeout protection

Output Location

• PDF Reports: ./Output/Summary/report.pdf
• Temp Audio: ./Output/Audio/ (auto-cleaned)

Notes

• Not a substitute for professional medical advice
• Designed for intake automation, not diagnosis
• Rate-limited API calls to prevent throttling
• Color-coded urgency in PDF (red for high-priority symptoms)