├── README.md
├── main.py
├── requirements.txt
├── sample.env
└── tools.py


/README.md:
--------------------------------------------------------------------------------
1 | # PythonAIAgentFromScratch


--------------------------------------------------------------------------------
/main.py:
--------------------------------------------------------------------------------
 1 | from dotenv import load_dotenv
 2 | from pydantic import BaseModel
 3 | from langchain_openai import ChatOpenAI
 4 | from langchain_anthropic import ChatAnthropic
 5 | from langchain_core.prompts import ChatPromptTemplate
 6 | from langchain_core.output_parsers import PydanticOutputParser
 7 | from langchain.agents import create_tool_calling_agent, AgentExecutor
 8 | from tools import search_tool, wiki_tool, save_tool
 9 | 
10 | load_dotenv()
11 | 
12 | class ResearchResponse(BaseModel):
13 |     topic: str
14 |     summary: str
15 |     sources: list[str]
16 |     tools_used: list[str]
17 |     
18 | 
19 | llm = ChatAnthropic(model="claude-3-5-sonnet-20241022")
20 | parser = PydanticOutputParser(pydantic_object=ResearchResponse)
21 | 
22 | prompt = ChatPromptTemplate.from_messages(
23 |     [
24 |         (
25 |             "system",
26 |             """
27 |             You are a research assistant that will help generate a research paper.
28 |             Answer the user query and use neccessary tools. 
29 |             Wrap the output in this format and provide no other text\n{format_instructions}
30 |             """,
31 |         ),
32 |         ("placeholder", "{chat_history}"),
33 |         ("human", "{query}"),
34 |         ("placeholder", "{agent_scratchpad}"),
35 |     ]
36 | ).partial(format_instructions=parser.get_format_instructions())
37 | 
38 | tools = [search_tool, wiki_tool, save_tool]
39 | agent = create_tool_calling_agent(
40 |     llm=llm,
41 |     prompt=prompt,
42 |     tools=tools
43 | )
44 | 
45 | agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True)
46 | query = input("What can i help you research? ")
47 | raw_response = agent_executor.invoke({"query": query})
48 | 
49 | try:
50 |     structured_response = parser.parse(raw_response.get("output")[0]["text"])
51 |     print(structured_response)
52 | except Exception as e:
53 |     print("Error parsing response", e, "Raw Response - ", raw_response)


--------------------------------------------------------------------------------
/requirements.txt:
--------------------------------------------------------------------------------
1 | langchain 
2 | wikipedia 
3 | langchain-community 
4 | langchain-openai
5 | langchain-anthropic
6 | python-dotenv
7 | pydantic
8 | duckduckgo-search


--------------------------------------------------------------------------------
/sample.env:
--------------------------------------------------------------------------------
1 | OPENAI_API_KEY=""
2 | ANTHROPIC_API_KEY=""


--------------------------------------------------------------------------------
/tools.py:
--------------------------------------------------------------------------------
 1 | from langchain_community.tools import WikipediaQueryRun, DuckDuckGoSearchRun
 2 | from langchain_community.utilities import WikipediaAPIWrapper
 3 | from langchain.tools import Tool
 4 | from datetime import datetime
 5 | 
 6 | def save_to_txt(data: str, filename: str = "research_output.txt"):
 7 |     timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
 8 |     formatted_text = f"--- Research Output ---\nTimestamp: {timestamp}\n\n{data}\n\n"
 9 | 
10 |     with open(filename, "a", encoding="utf-8") as f:
11 |         f.write(formatted_text)
12 |     
13 |     return f"Data successfully saved to {filename}"
14 | 
15 | save_tool = Tool(
16 |     name="save_text_to_file",
17 |     func=save_to_txt,
18 |     description="Saves structured research data to a text file.",
19 | )
20 | 
21 | search = DuckDuckGoSearchRun()
22 | search_tool = Tool(
23 |     name="search",
24 |     func=search.run,
25 |     description="Search the web for information",
26 | )
27 | 
28 | api_wrapper = WikipediaAPIWrapper(top_k_results=1, doc_content_chars_max=100)
29 | wiki_tool = WikipediaQueryRun(api_wrapper=api_wrapper)
30 | 
31 | 


--------------------------------------------------------------------------------