search

Overview

This project is a web search Q&A system that uses natural language processing (NLP) to answer user queries based on search results from the web. The system uses LLaMA 3 70B to extract content from web pages and generate answers based on the search results and/or an instruction.

Main Features

• Web Search Q&A
• Query Type Evaluation
• Math Query Handling
• Simple Query Handling
• Best Answer Selection

Requirements

• Python 3.8+
• requests library for making HTTP requests
• BeautifulSoup library for parsing HTML content
• Groq API key for answer generation

Server Usage Examples

1. Web Search Q&A (Default)

• Query: What is the capital of France?
  • The system will perform a web search and provide an answer based on the results.

2. Math Queries

• Query: What is 15 * 7 + 3?
  • The system will detect this as a math query and provide the calculated result.

3. Simple Queries

• Query: Who wrote "To Kill a Mockingbird"?
  • For well-known information, the system may provide a direct answer without web search.

4. Realtime Information

• Query: What's the current weather in New York?
  • The system will recognize this as requiring up-to-date information and perform a web search.
• Best Answer Selection
  • For complex queries, the system may generate multiple answers and select the best one.

API Usage

Send a POST request to /ask endpoint with JSON payload:

{
  "query": "Your question here",
  "max_retries": 3  // Optional, default is 3
}

Response format:

{
  "response": "Answer to the query",
  "type": "search" // or "math" or "text"
}

Note: The system automatically determines the query type and processes accordingly.

License

This project is licensed under the MIT License. See the LICENSE file for more information.