|
|
import gradio as gr |
|
|
|
|
|
from typing import TypedDict, Annotated, Optional |
|
|
|
|
|
import requests |
|
|
from bs4 import BeautifulSoup |
|
|
from dotenv import load_dotenv |
|
|
from pydantic import BaseModel, Field |
|
|
|
|
|
from runnable_with_tools import RunnableWithTools |
|
|
from langchain_anthropic import ChatAnthropic |
|
|
from langchain_core.prompts import ChatPromptTemplate |
|
|
from langchain_core.tools import tool |
|
|
|
|
|
class WebPage(BaseModel): |
|
|
""" Information about a web page, including its title, content and useful links """ |
|
|
url: str = Field(description="URL of this web page") |
|
|
title: Optional[str] = Field(default=None, description="The web page title") |
|
|
content: Optional[str] = Field(default=None, description="Web page content") |
|
|
links: list[str] = Field(default_factory=list, description="the useful links on this web page") |
|
|
|
|
|
@tool |
|
|
def get_web_page(url: Annotated[str, 'the url of the web page']) -> Annotated[WebPage, 'information on the web page']: |
|
|
""" Retrieve information about a web page, including its title, content and useful links """ |
|
|
print(f'Retrieving {url}') |
|
|
response = requests.get(url) |
|
|
soup = BeautifulSoup(response.content, 'html.parser') |
|
|
title = soup.title.string if soup.title else None |
|
|
content = None |
|
|
if soup.body: |
|
|
for irrelevant in soup.body(["script", "style", "img", "input"]): |
|
|
irrelevant.decompose() |
|
|
content = soup.body.get_text(separator="\n", strip=True) |
|
|
links = [link.get('href') for link in soup.find_all('a') if link.get('href') is not None] |
|
|
return WebPage(url=url, title=title, content=content, links=links) |
|
|
|
|
|
TOOLS = { |
|
|
"get_web_page": get_web_page |
|
|
} |
|
|
|
|
|
def make_brochure(company_name: str, company_web_site: str) -> str: |
|
|
print(f'processing request for {company_name} at {company_web_site}') |
|
|
|
|
|
system = """ |
|
|
You are an assistant that can crate company brochure from company's web site. |
|
|
You can do this in a few steps: |
|
|
1. Retrieve web page information from the provide company web site url |
|
|
2. Find out the links that are useful for making company brochure |
|
|
3. Retrieve web page information from these links |
|
|
4. Create a company brochure from information on these wwn pages |
|
|
""" |
|
|
|
|
|
user = """ |
|
|
Please create a company brochure for {company_name}. Its web site url is {company_web_site} |
|
|
""" |
|
|
|
|
|
prompt_template = ChatPromptTemplate.from_messages([ |
|
|
('system', system), |
|
|
('user', user) |
|
|
]) |
|
|
prompt = prompt_template.invoke({'company_name': company_name, 'company_web_site': company_web_site}) |
|
|
brochure = model_with_tools.invoke(prompt.to_messages(), max_depth=20).content |
|
|
print(f'processed request for {company_name} at {company_web_site}') |
|
|
print(f'brochure is: \n') |
|
|
print(brochure) |
|
|
return brochure |
|
|
|
|
|
load_dotenv() |
|
|
model = ChatAnthropic(model="claude-3-5-haiku-20241022", temperature=0) |
|
|
model_with_tools = RunnableWithTools(bound=model.bind_tools([get_web_page]), tools=TOOLS) |
|
|
|
|
|
app = gr.Interface(fn=make_brochure, |
|
|
inputs=[gr.Textbox(label='Company name'), gr.Textbox(label='Company web site')], |
|
|
outputs=gr.Markdown(label='Company brochure', value='# Company brochure', show_label=True, container=True), |
|
|
flagging_mode="never") |
|
|
app.launch() |
