Skip to content

README_web_app.md

Latest commit

 

History

History
123 lines (102 loc) · 3.02 KB

README_web_app.md

File metadata and controls

123 lines (102 loc) · 3.02 KB
Raw

🌐 Government AI Assistant Web Application

Overview

H.A.R.V.E.Y Web App is an interactive interface for querying government policies using AI. Focused exclusively on the web components:

  • FastAPI Backend: REST API with CORS and async processing
  • Streamlit Frontend: Stateful chat interface with real-time updates
  • Configuration Management: JSON-based endpoint configuration
  • Production-Ready: Port management and service orchestration

🖥️ Web-Specific Requirements

  • Core Services: 
    fastapi 
uvicorn
streamlit
aiohttp
  • Network: Localhost ports 8000 (API), 8501 (UI), 8001 (MLflow)

🚀 Web App Setup Guide

1. Install Dependencies

pip install -r src/requirements.txt

2. Backend Initialization

# Start services (MLflow + FastAPI) from project root
python src/backend/services.py

View MLFlow

MLFLOW can be viewed below link after this log - INFO: Application startup complete.

MLFLOW Link - http://0.0.0.0:8001

3. Frontend Activation

streamlit run src/frontend/streamlit.py --server.port 8501

View Frontend

Frontend can be viewed on this link - http://localhost:8501

🔄 API-Frontend Integration

Endpoint Configuration

Modify frontend_config.json for different environments:

{
  "CHAT_API": {
    "endpoint": "http://your-production-domain:8000",
    "chat_response_route": "/answer_query"
  }
}

Session Management Flow

sequenceDiagram
    participant User
    participant Streamlit
    participant FastAPI
    User->>Streamlit: Enters query
    Streamlit->>FastAPI: POST /answer_query
    FastAPI->>Streamlit: {answer, metadata}
    Streamlit->>User: Displays formatted response
Loading

🛠️ Key Web Components

Backend Services (services.py)

def start_mlflow():
    # Starts MLflow tracking server
    subprocess.Popen(["mlflow", "server", ...])

def start_fastapi():
    # Launches ASGI server
    uvicorn.run(...)

Chat Endpoint (llm_routes.py)

@answer_query_router.post("/")
async def answer_query(query_body: MessagesList):
    # Processes 3 types of payloads:
    # 1. New user queries
    # 2. Follow-up questions
    # 3. Contextual requests

Frontend State Management

# Session state initialization
if "messages" not in st.session_state:
    st.session_state.messages = []

# Message processing
async def query_api(messages, api_url):
    # Handles:
    # - Timeouts (10s limit)
    # - JSON serialization
    # - Error recovery

📂 Web Component Structure

web-app/
├── backend/
│   ├── services.py       # Service orchestrator
│   ├── backend.py        # Core API config
│   ├── routes/           # API endpoints
│   └── models/           # Request/response schemas
└── frontend/
    ├── streamlit.py      # Chat UI core
    ├── frontend.py       # UI launcher
    └── frontend_config.json

For RAG pipeline documentation, see RAG_README.md in the main directory.