The LangGraph Reflection<\/a> Framework\u00a0is a type of agentic framework which offers a powerful way to improve language model outputs through an iterative critique process using Generative AI. This article breaks down how to implement a reflection agent that validates Python code using Pyright and improves its quality using GPT-4o mini<\/a>.\u00a0AI agents play a crucial role in this framework, automating decision-making processes by combining reasoning, reflection, and feedback mechanisms to enhance model performance.<\/p>\n This article was published as a part of the\u00a0<\/strong><\/em>Data Science Blogathon.<\/strong><\/em><\/a><\/p>\n The LangGraph Reflection Framework follows a simple yet effective agentic architecture:<\/p>\n Also Read: Agentic Frameworks for Generative AI Applications<\/a><\/em><\/p>\n Here is a Step-by-Step Guide<\/b> for an Illustrative Implementation and Usage:<\/p>\n First, install the required dependencies:<\/p>\n We\u2019ll use Pyright to analyze generated code and provide error details.<\/p>\n Pyright Analysis Function<\/strong><\/p>\n GPT-4o Mini Model Setup<\/b><\/p>\n Note: Use os.environ[\u201cOPENAI_API_KEY\u201d] = \u201cYOUR_API_KEY<\/i>\u201d securely, and never hardcode the key in your code.<\/p>\n Code Extraction Types<\/b><\/p>\n System Prompt for GPT-4o Mini<\/b><\/p>\n Pyright Code Validation Function<\/b><\/p>\n Building the Main and Judge Graphs<\/b><\/p>\n Example Execution<\/b><\/p>\n Our LangGraph Reflection system was designed to do the following:<\/p>\n 1. Import \u201cfaiss\u201d could not be resolved.<\/p>\n 2. Cannot access attribute \u201cembed\u201d for class \u201cOpenAIEmbeddings\u201d.<\/p>\n 3. Arguments missing for parameters \u201cdocstore\u201d, \u201cindex_to_docstore_id\u201d.<\/p>\n In the second iteration, the system improved the code but still identified:<\/p>\n 1. Import \u201clangchain.document\u201d could not be resolved.<\/p>\n 2. \u201cInMemoryDocstore\u201d is not defined.<\/p>\n In the final iteration, the reflection agent successfully addressed all issues by:<\/p>\n The improved code then successfully ran without errors.<\/p>\n The LangGraph Reflection Framework demonstrates the power of combining AI critique agents with robust static analysis tools. This intelligent feedback loop enables faster code correction, improved coding practices, and better overall development efficiency. Whether for beginners or experienced developers, LangGraph Reflection offers a powerful tool for improving code quality.<\/p>\n The media shown in this article is not owned by Analytics Vidhya and is used at the Author\u2019s discretion.<\/strong><\/p>\n A. LangGraph Reflection is a powerful framework that combines a primary AI agent (for code generation or task execution) with a critique agent (to identify issues and suggest improvements). This iterative loop improves the final output by leveraging feedback and reflection.<\/p>\n<\/p><\/div>\n A. The reflection mechanism follows this workflow: A. You\u2019ll need the following dependencies: To install them, run:\u00a0pip install langgraph-reflection langchain pyright faiss openai<\/p>\n<\/p><\/div>\n A. LangGraph Reflection excels at tasks like: A. No, while we have shown Pyright in code correction examples, the framework can even help in improving text summarization, data validation, and chatbot response refinement.<\/p>\n<\/p><\/div>\n<\/p><\/div>\nLearning Objectives<\/h4>\n
\n
LangGraph Reflection Framework Architecture<\/h2>\n
\n
![\"LangGraph](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/03\/langgraph-reflection.webp\")
<\/figure>\n<\/div>\nHow to Implement the LangGraph Reflection Framework<\/h2>\n
Step 1: Environment Setup<\/h3>\n
pip install langgraph-reflection langchain pyright<\/code><\/pre>\nStep 2: Code Analysis with Pyright<\/h3>\n
from typing import TypedDict, Annotated, Literal\nimport json\nimport os\nimport subprocess\nimport tempfile\n\nfrom langchain.chat_models import init_chat_model\nfrom langgraph.graph import StateGraph, MessagesState, START, END\nfrom langgraph_reflection import create_reflection_graph\n\nos.environ[\"OPENAI_API_KEY\"] = \"your_openai_api_key\"\n\ndef analyze_with_pyright(code_string: str) -> dict:\n \"\"\"Analyze Python code using Pyright for static type checking and errors.\n\n Args:\n code_string: The Python code to analyze as a string\n\n Returns:\n dict: The Pyright analysis results\n \"\"\"\n with tempfile.NamedTemporaryFile(suffix=\".py\", mode=\"w\", delete=False) as temp:\n temp.write(code_string)\n temp_path = temp.name\n\n try:\n result = subprocess.run(\n [\n \"pyright\",\n \"--outputjson\",\n \"--level\",\n \"error\", # Only report errors, not warnings\n temp_path,\n ],\n capture_output=True,\n text=True,\n )\n\n try:\n return json.loads(result.stdout)\n except json.JSONDecodeError:\n return {\n \"error\": \"Failed to parse Pyright output\",\n \"raw_output\": result.stdout,\n }\n finally:\n os.unlink(temp_path)<\/code><\/pre>\nStep 3: Main Assistant Model for Code Generation<\/h3>\n
def call_model(state: dict) -> dict:\n \"\"\"Process the user query with the GPT-4o mini model.\n\n Args:\n state: The current conversation state\n\n Returns:\n dict: Updated state with the model response\n \"\"\"\n model = init_chat_model(model=\"gpt-4o-mini\", openai_api_key = 'your_openai_api_key')\n return {\"messages\": model.invoke(state[\"messages\"])}<\/code><\/pre>\nStep 4: Code Extraction and Validation<\/h3>\n
# Define type classes for code extraction\nclass ExtractPythonCode(TypedDict):\n \"\"\"Type class for extracting Python code. The python_code field is the code to be extracted.\"\"\"\n python_code: str\n\n\nclass NoCode(TypedDict):\n \"\"\"Type class for indicating no code was found.\"\"\"\n no_code: bool<\/code><\/pre>\n# System prompt for the model\nSYSTEM_PROMPT = \"\"\"The below conversation is you conversing with a user to write some python code. Your final response is the last message in the list.\n\nSometimes you will respond with code, othertimes with a question.\n\nIf there is code - extract it into a single python script using ExtractPythonCode.\n\nIf there is no code to extract - call NoCode.\"\"\"<\/code><\/pre>\ndef try_running(state: dict) -> dict | None:\n \"\"\"Attempt to run and analyze the extracted Python code.\n\n Args:\n state: The current conversation state\n\n Returns:\n dict | None: Updated state with analysis results if code was found\n \"\"\"\n model = init_chat_model(model=\"gpt-4o-mini\")\n extraction = model.bind_tools([ExtractPythonCode, NoCode])\n er = extraction.invoke(\n [{\"role\": \"system\", \"content\": SYSTEM_PROMPT}] + state[\"messages\"]\n )\n if len(er.tool_calls) == 0:\n return None\n tc = er.tool_calls[0]\n if tc[\"name\"] != \"ExtractPythonCode\":\n return None\n\n result = analyze_with_pyright(tc[\"args\"][\"python_code\"])\n print(result)\n explanation = result[\"generalDiagnostics\"]\n\n if result[\"summary\"][\"errorCount\"]:\n return {\n \"messages\": [\n {\n \"role\": \"user\",\n \"content\": f\"I ran pyright and found this: {explanation}\\n\\n\"\n \"Try to fix it. Make sure to regenerate the entire code snippet. \"\n \"If you are not sure what is wrong, or think there is a mistake, \"\n \"you can ask me a question rather than generating code\",\n }\n ]\n }<\/code><\/pre>\nStep 5: Creating the Reflection Graph<\/h3>\n
def create_graphs():\n \"\"\"Create and configure the assistant and judge graphs.\"\"\"\n # Define the main assistant graph\n assistant_graph = (\n StateGraph(MessagesState)\n .add_node(call_model)\n .add_edge(START, \"call_model\")\n .add_edge(\"call_model\", END)\n .compile()\n )\n\n # Define the judge graph for code analysis\n judge_graph = (\n StateGraph(MessagesState)\n .add_node(try_running)\n .add_edge(START, \"try_running\")\n .add_edge(\"try_running\", END)\n .compile()\n )\n\n # Create the complete reflection graph\n return create_reflection_graph(assistant_graph, judge_graph).compile()\n\nreflection_app = create_graphs()<\/code><\/pre>\nStep 6: Running the Application<\/h3>\n
if __name__ == \"__main__\":\n \"\"\"Run an example query through the reflection system.\"\"\"\n example_query = [\n {\n \"role\": \"user\",\n \"content\": \"Write a LangGraph RAG app\",\n }\n ]\n\n print(\"Running example with reflection using GPT-4o mini...\")\n result = reflection_app.invoke({\"messages\": example_query})\n print(\"Result:\", result)<\/code><\/pre>\nOutput Analysis<\/h3>\n
![\"enhanced](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/03\/image_DErfJhL.webp\")
<\/figure>\n![\"How](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/03\/download_VD0VXh6.png\")
<\/figure>\nWhat Happened in the Example?<\/h2>\n
\n
Iteration 1 \u2013 Identified Errors<\/h3>\n
\n
pip install faiss-cpu<\/code><\/pre>\n\n
\n
Iteration 2 \u2013 Progression<\/h3>\n
\n
\n
from langchain.docstore import InMemoryDocstore<\/code><\/pre>\nIteration 3 \u2013 Final Solution<\/h3>\n
\n
Why This Matters<\/h4>\n
\n
Conclusion<\/h2>\n
Key Takeaways<\/h4>\n
\n
Frequently Asked Questions<\/h2>\n
\u2013 Main Agent:<\/strong> Generates the initial output.
\u2013 Critique Agent:<\/strong> Analyzes the generated output for errors or improvements.
\u2013 Improvement Loop: <\/strong>If issues are found, the main agent is re-invoked with feedback for refinement. This loop continues until the output meets quality standards.<\/p>\n<\/p><\/div>\n
\u2013 langgraph-reflection
\u2013 langchain
\u2013 pyright (for code analysis)
\u2013 faiss (for vector search)
\u2013 openai (for GPT-based models)<\/p>\n
\u2013 Python code validation and improvement.
\u2013 Natural language responses requiring fact-checking.
\u2013 Document summarization with clarity and completeness.
\u2013 Ensuring AI-generated content adheres to safety guidelines.<\/p>\n<\/p><\/div>\n
![\"Adarsh](\"https:\/\/av-eks-lekhak.s3.amazonaws.com\/media\/lekhak-profile-images\/converted_image_tHXFGNS.webp\")
<\/p>\n