blog_evaluater_optimizer.py

import os
from dotenv import load_dotenv
from typing import Literal, List, Dict, TypedDict
from langchain_groq import ChatGroq
from pydantic import BaseModel, Field
from langsmith import traceable
from langgraph.graph import StateGraph, START, END
from IPython.display import Image, display

load_dotenv()

os.environ["GROQ_API_KEY"] = os.getenv("GROQ_API_KEY")
os.environ["LANGCHAIN_TRACING_V2"] = "true"
os.environ["LANGCHAIN_API_KEY"] = os.getenv("LANGCHAIN_API_KEY")

llm = ChatGroq(model="qwen-2.5-32b")

# Graph state
class State(TypedDict):
    blog: str
    topic: str
    feedback: str
    good_or_revise: str


class Feedback(BaseModel):
    grade: Literal["good", "needs revision"] = Field(
        description="Decide if the blog is entertaining, concise with maxiumum of 400 characters, with subtitles and a conclusion or needs revision.",
    )
    feedback: str = Field(
        description="If the blog is not good, provide feedback on how to improve it.",
    )


evaluator = llm.with_structured_output(Feedback)


# Nodes
@traceable
def llm_call_generator(state: State):
    """LLM generates a blog"""
    if state.get("feedback"):
        msg = llm.invoke(
            f"Write a blog about {state['topic']} but take into account the feedback: {state['feedback']}"
        )
    else:
        msg = llm.invoke(f"Write a blog about {state['topic']}")
    
    # Debugging print statement
    print("Generated blog content:", msg.content)  
    
    return {"blog": msg.content}  # Ensure this key is returned!


@traceable
def llm_call_evaluator(state: State):
    """LLM evaluates the blog"""
    grade = evaluator.invoke(f"Grade the blog {state['blog']}")
    return {"good_or_revise": grade.grade, "feedback": grade.feedback}


@traceable
def route_blog(state: State):
    """Route back to blog generator or end based upon feedback from evaluator"""
    if state["good_or_revise"] == "good":
        return "Accepted"
    elif state["good_or_revise"] == "needs revision":
        return "llm_call_generator"


# Build workflow
optimizer_builder = StateGraph(State)

# Add the nodes
optimizer_builder.add_node("llm_call_generator", llm_call_generator)
optimizer_builder.add_node("llm_call_evaluator", llm_call_evaluator)

# Add edges to connect nodes
optimizer_builder.add_edge(START, "llm_call_generator")
optimizer_builder.add_edge("llm_call_generator", "llm_call_evaluator")
optimizer_builder.add_conditional_edges(
    "llm_call_evaluator",
    route_blog,
    {
        "Accepted": END,
        "llm_call_generator": "llm_call_generator",
    },
)

# Compile the workflow
optimizer_workflow = optimizer_builder.compile()

# Show the workflow
display(Image(optimizer_workflow.get_graph().draw_mermaid_png()))

# Invoke
state = optimizer_workflow.invoke({"topic": "Vibe Coding"})
print(state["blog"])