market_research_agent_web.py

import gradio as gr
import os
import requests
from bs4 import BeautifulSoup
import cohere
from dotenv import load_dotenv

# Load environment variables
load_dotenv()
COHERE_API_KEY = os.getenv("COHERE_API_KEY")
co = cohere.Client(COHERE_API_KEY)

# DuckDuckGo search function
def web_search_duckduckgo(query, max_results=5):
    DUCKDUCKGO_SEARCH_URL = "https://html.duckduckgo.com/html/"
    HEADERS = {"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36"}
    
    resp = requests.post(DUCKDUCKGO_SEARCH_URL, data={"q": query}, headers=HEADERS)
    soup = BeautifulSoup(resp.text, "html.parser")
    links = soup.find_all("a", class_="result__a", limit=max_results)
    
    web_docs = []
    for link in links:
        href = link["href"]
        page = requests.get(href)
        text = BeautifulSoup(page.text, "html.parser").get_text()
        web_docs.append(text)
    return web_docs

# Function to query Cohere for reasoning
def agentic_rag_with_cohere(query, web_docs):
    formatted_docs = [{"text": doc} for doc in web_docs]
    response = co.chat(
        model="command-r-plus",
        message=query,
        documents=formatted_docs,
    )
    return response.text

# Gradio Interface function
def market_research_agent(query):
    web_docs = web_search_duckduckgo(query)
    response = agentic_rag_with_cohere(query, web_docs)
    return response

# Create the Gradio Interface
iface = gr.Interface(
    fn=market_research_agent, 
    inputs=gr.Textbox(label="Enter Your Market Research Query"), 
    outputs=gr.Textbox(label="Agent Response"),
    title="Market Research Agent",
    description="This agent uses real-time web search and Cohere's LLM for adaptive market research."
)

# Launch the Gradio interface
iface.launch()