README.md

metadata

title: Help.me AI Operator
emoji: 🚑
colorFrom: blue
colorTo: purple
sdk: docker
app_port: 7860
pinned: false

<<<<<<< HEAD

title: AI Operator emoji: 🦀 colorFrom: pink colorTo: yellow sdk: docker pinned: false license: mit short_description: 'Agentic AI which is specialised answering emergency calls '

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

Help.me - AI-Powered Emergency Medical Assistance System

"Help.me" is an AI-based platform designed to automate and accelerate communication between patients in need of urgent medical care and dispatch operators. The system receives voice messages from patients, analyzes their condition using artificial intelligence, and immediately transmits the information to a dispatcher dashboard for appropriate action.

🚀 Key Features

For Patients (Voice-First Interface):

Voice Communication: Patients can report their symptoms simply by speaking.

Multilingual System: The AI can communicate with patients in Uzbek, Russian, and English.

Smart Recommendations: If a patient's condition is assessed as "Green" (non-urgent), the system recommends public polyclinics or private clinics based on their symptoms.

Real-time Response: The AI analyzes the request instantly and provides a voice response.

Simplified Interface: The interface is designed to be as simple as possible and voice-focused to avoid distracting the patient in stressful situations.

For Dispatchers (Monitoring Dashboard):

Real-time Monitoring: All incoming cases are displayed live on the dashboard.

Risk Triage: The AI categorizes each case into "Red" (emergency), "Yellow" (uncertain), or "Green" (clinic referral) risk levels.

Interactive Map: The locations of all ambulance brigades and clinics are tracked on a map in real time.

Statistics & Analytics: Statistical data on cases and brigades are visualized in charts.

🧠 AI Models Used

Our system relies on three core AI models:

Speech-to-Text (STT):

Model: A custom model fine-tuned on top of OpenAI Whisper (medium).

Dataset: The model was trained on several datasets tailored to the conditions of Uzbekistan. This includes audio recordings in the Tashkent dialect, standard literary language, and additionally, the Khorezm dialect. This ensures high accuracy in understanding the speech of patients from various regions of the country.

Logic and Response Generation (LLM):

Model: Google Gemini Flash.

Task: To analyze the transcribed complaints from the patient, determine the severity of the situation (risk level), and formulate the text of the response. The model is guided by a strict set of rules and action sequences provided via a SYSTEM_INSTRUCTION.

Text-to-Speech (TTS):

Model: Facebook MMS (Massively Multilingual Speech).

Task: To synthesize the AI-generated response text into a natural-sounding human voice. The system uses separate TTS models for Uzbek and English.

🛠️ Technology Stack

Backend: FastAPI (Python)

Real-time Communication: WebSockets

Database: JSON-based flat files (for MVP)

Frontend: HTML, CSS, JavaScript (Vanilla JS)

Map: Leaflet.js, Charts: Chart.js

⚙️ Getting Started

1. Prerequisites:

Python 3.9+

FFmpeg (must be installed on the system to process audio files)

git

3. Set Up Virtual Environment and Install Dependencies:

Create a virtual environment

python -m venv venv

Activate it (Windows)

venv\Scripts\activate

Activate it (MacOS/Linux)

source venv/bin/activate

Install the required libraries

pip install -r requirements.txt

4. ‼️ IMPORTANT: Download AI Models

This repository DOES NOT include the large AI models in the local_models directory. To run the system, you must download them separately and place them in the project folder.

Note for the judges: Due to their large size (several GB), it was not feasible to upload the models to GitHub. The repository contains only the project's source code.

5. Run the Application:

uvicorn app.main:app --reload

0f59686 (Loyiha tayyor: Help.me AI tizimi)