Initial commit: Multi-Agent research pipeline using LangGraph and Gemini

.gitignore

@@ -0,0 +1,4 @@
+venv/
+__pycache__/
+.env
+*.pyc

agents.py

@@ -0,0 +1,79 @@
+# This is the heart of the project. We will build 4 things here. First the Search Agent using create_react_agent + AgentExecutor which will use the web_search tool. Second the Reader Agent using the same pattern but with the scrape_url tool. Third the Write Chain using the modern LCEL pipe syntax -prompt | 11m | StrOutputParser() which takes the research and writes a full report. Fourth the Critic CHain again using LCEL pipeline which reads the report and gives a score and feedback
+from langchain.agents import create_agent
+from langchain_google_genai import ChatGoogleGenerativeAI
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_core.output_parsers import StrOutputParser
+from tools import web_search, scrape_url
+from dotenv import load_dotenv  
+
+load_dotenv()  # Load environment variables from .env file
+
+#Model Setup
+llm= ChatGoogleGenerativeAI(model="gemini-2.5-flash", temperature=0)
+
+
+#1st Agent
+def build_search_agent():
+    return create_agent(
+        model = llm,
+        tools= [web_search],
+        system_prompt="You are a search agent. You must search the web to find recent and reliable information. Always use the web_search tool to find URLs and actual sources. Return the search results."
+    )
+
+
+#2nd Agent
+def build_reader_agent():
+    return create_agent(
+        model= llm,
+        tools= [scrape_url],
+        system_prompt="You are a reader agent. You must select the most relevant URL from the search results and use the scrape_url tool to extract its content. Always perform scraping using the tool."
+    )
+
+#Writer chain
+
+writer_prompt= ChatPromptTemplate.from_messages([
+    ('system', "You are an expert research writer. Write clear, structured and insightful reports. "),
+    ('human', """Write a detailed research report on the topic below.
+     Topic: {topic} 
+     
+     Research Gathered:
+     {research}
+     
+     Structure the report as:
+     -Introduction
+     -Key Findings (minimum 3 well-explained points)
+     -Conclusion
+     -Sources (list all URLs found in the research)
+
+     Be detsailed, factual and professional."""),
+])
+
+writer_chain= writer_prompt | llm | StrOutputParser()
+
+#Critic Chain
+
+critic_prompt= ChatPromptTemplate.from_messages([
+ ('system', "You are an expert research writer. Write clear, structured and insightful reports. "),
+ ('human', """Write a detailed research report on the topic below.
+  
+  Report:
+  {report}
+  
+  Respond in this exact format:
+  
+  Score : X/10
+  
+  Strengths:
+   - ...
+   - ...
+  
+  Areas to Improve:
+  - ...
+  - ...
+  
+  One liner verdict:
+  ..."""),
+        
+])
+
+critic_chain= critic_prompt | llm | StrOutputParser()

pipeline.py

@@ -0,0 +1,80 @@
+from agents import build_reader_agent, build_search_agent, writer_chain, critic_chain
+
+def extract_text(content) -> str:
+    if isinstance(content, str):
+        return content
+    elif isinstance(content, list):
+        return "".join(item.get("text", "") for item in content if isinstance(item, dict) and "text" in item)
+    return str(content)
+
+def run_research_pipeline(topic: str) -> dict:
+    
+    state={}
+
+    #Search Agent working
+
+    print("\n" + "="*50 )
+    print("step 1 - search agent is wokring ...")
+    print("=" *50)
+
+    search_agent= build_search_agent()
+    search_result= search_agent.invoke({ 
+        "messages": [("user", f"Find recent, reliable and detailed information about: {topic}")]
+    })
+
+    state["search_results"]= extract_text(search_result['messages'][-1].content)
+    print("\n search result", state['search_results'])
+
+    #Step 2 - reader agent
+    print("\n" + "="*50)
+    print("step 2- reader agent is scrapping top respurces ...")
+    print("="*50)
+
+    reader_agent= build_reader_agent()
+    reader_result = reader_agent.invoke({ 
+        "messages": [("user",
+        f"Based on the following search results about '{topic}',"
+        f"pick the most relevant URL and scrape it for deeper content.\n\n"
+        f"Search Results: \n{state['search_results'][:800]}"
+        )]
+    }) 
+
+    state['scraped_content']= extract_text(reader_result['messages'][-1].content)
+    print("\nScraped content\n", state['scraped_content'])
+
+    #Step 3- writer chain
+
+    print("\n" + "="*50)
+    print("step 3- Writer is drafting the report ...")
+    print("="*50)
+
+    research_combined= ( 
+        f"Search Results: \n {state['search_results']}\n\n"
+        f"Detailed Scraped Content: \n {state['scraped_content']}"
+    
+    )
+
+    state['report']= writer_chain.invoke({
+        "topic":topic,
+        "research": research_combined
+    })
+
+    print("\n final report\n", state['report'])
+
+    #Critic Report
+    print("\n" + "="*50)
+    print("step 3- Critic is reviewing the report ...")
+    print("="*50)
+
+    state['feedback']=critic_chain.invoke({
+        "report": state['report']
+    })
+
+    print("\n critic report \n", state['feedback'])
+
+    return state
+
+if __name__ == "__main__":
+    topic= input("\n Enter a research topic: " )
+    run_research_pipeline(topic)
+    

requirement.txt

@@ -0,0 +1,23 @@
+langchain
+langchain-core
+langchain-community
+langgraph
+langchain-google-genai>=2.0.0
+
+tavily-python
+beautifulsoup4
+requests
+lxml
+
+python-dotenv
+pydantic
+httpx
+aiohttp
+tiktoken
+tenacity
+
+fastapi
+uvicorn
+
+langsmith
+pytest

tools.py

@@ -0,0 +1,40 @@
+# We build 2 custom tools using the @tool decorator. Firest the web_search tool which talks to the Tavily API and fetches live search results from the internet. Second the scrape_url tool which takes a URL, visits that page and extracts clean and readable text from it using BeautifulSoup.
+from langchain.tools import tool
+import requests
+from bs4 import BeautifulSoup
+from tavily import TavilyClient
+import os
+from dotenv import load_dotenv
+from rich import print
+
+load_dotenv()
+
+tavily = TavilyClient(api_key=os.getenv("TAVILY_API_KEY"))
+@tool
+def web_search(query : str ) -> str:
+    """Search the web for recent and reliable information on a given topic. Returns Titles, URLs and snippets """
+    results = tavily.search(query= query, max_result=5)
+    out=[]
+
+    for r in results['results']:
+        out.append(
+            f'Title: {r["title"]}\nURL: {r["url"]}\nSnippet: {r["content"][:300]}\n'
+        )
+    return "\n----\n".join(out)
+        
+# print(web_search.invoke("What is the recent news of war"))
+
+
+@tool
+def scrape_url(url: str) -> str:
+    """Scrape and return clean text content from a given URL."""
+    try:
+        resp= requests.get(url, timeout=8, headers= {"User-Agent": "Mozilla/5.0"})
+        soup= BeautifulSoup(resp.text, "html.parser")
+        for tag in soup(["script","style", "nav", "footer"]):
+            tag.decompose()
+        return soup.get_text(separator="\n", strip=True)[:3000]
+    except Exception as e:
+        return f"Could not scrape URL: {str(e)}"
+    
+# print(scrape_url.invoke("https://www.bbc.com/news/world-europe-66707497"))