Agentic Design Patterns

This repository contains a small set of Python examples that explore agentic design patterns with:

• LangChain + OpenAI
• Google ADK + Gemini

The code is organized as a progression from simple sequential chains to coordinator, parallel, reflection, and review-based agent workflows.

Repository Contents

Code_1.py

A simple sequential LangChain LCEL example using OpenAI.

What it does:

• extracts technical specifications from free text
• transforms the extracted information into a JSON-like structure with cpu, memory, and storage

Key libraries:

• langchain-openai
• langchain-core

API key:

• OPENAI_API_KEY

Run:

python Code_1.py

Code_2.py

A LangChain coordinator-style routing example using OpenAI.

What it does:

• classifies a request into booking or information handling
• routes the request through a coordinator chain
• delegates to simulated Python handlers
• prints the final delegated result

Key libraries:

• langchain-openai
• langchain-core

API key:

• OPENAI_API_KEY

Run:

python Code_2.py

Code_3.py

A Google ADK coordinator example using Gemini.

What it does:

• defines a coordinator agent plus specialized sub-agents
• wraps Python functions as ADK tools
• creates an in-memory session and runner
• delegates booking or information requests through ADK
• prints the final response text

Key libraries:

• google-adk
• google-genai

API key:

• GOOGLE_API_KEY or GEMINI_API_KEY

Run:

python Code_3.py

Notes:

• this example depends on Gemini quota and billing availability
• if you see 429 RESOURCE_EXHAUSTED, the issue is usually project quota rather than Python code
• you may see warnings about function_call parts in the returned content; those indicate tool usage, not necessarily failure

Code_4.py

A LangChain parallel-processing example using OpenAI.

What it does:

• runs three chains in parallel for the same topic
• produces:
  • a summary
  • a list of interesting questions
  • a set of key terms
• synthesizes the parallel outputs into a single final response

Key libraries:

• langchain-openai
• langchain-core

API key:

• OPENAI_API_KEY

Run:

python Code_4.py

Code_5.py

A Google ADK parallel-agent example using Gemini and Google Search.

What it does:

• creates three research agents that run in parallel
• each sub-agent researches one sustainability topic:
  • renewable energy
  • electric vehicles
  • carbon capture
• stores each result in shared session state
• runs a synthesis agent after the parallel stage
• prints the final structured report

Key libraries:

• google-adk
• google-genai

API key:

• GOOGLE_API_KEY or GEMINI_API_KEY

Run:

python Code_5.py

Notes:

• this example uses the Google Search tool through ADK
• because it is a multi-step parallel workflow, runtime depends on model/tool availability and Gemini quota

Code_6.py

A LangChain reflection-loop example using OpenAI.

What it does:

• asks the model to write a Python calculate_factorial function
• runs an iterative generate-and-critique loop
• uses a reviewer prompt to inspect the current code against the original requirements
• stops early if the reviewer returns CODE_IS_PERFECT
• prints the refined code after the loop completes

Key libraries:

• langchain-openai
• langchain-core
• python-dotenv

API key:

• OPENAI_API_KEY

Run:

python Code_6.py

Notes:

• this example loads environment variables from a local .env file if present
• it demonstrates a reflection pattern rather than tool-based multi-agent orchestration

Code_7.py

A Google ADK draft-and-review pipeline example using Gemini.

What it does:

• creates a draft-writing agent and a reviewer agent
• runs both agents sequentially through an ADK SequentialAgent
• stores the generated paragraph in shared state as draft_text
• passes that draft to a reviewer that returns a structured critique in review_output
• executes the pipeline end to end with an in-memory runner

Key libraries:

• google-adk
• google-genai

API key:

• GOOGLE_API_KEY or GEMINI_API_KEY

Run:

python Code_7.py

Notes:

• this example demonstrates sequential draft generation followed by critique
• the sample request uses quantum computing to make the review stage more demanding

Code_8.py

A LangChain tool-calling example using Gemini.

What it does:

• defines a simulated information lookup tool
• creates a Gemini-backed LangChain tool-calling agent
• asks several questions concurrently with asyncio.gather
• prints the tool calls and final agent responses

Key libraries:

• langchain-google-genai
• langchain-core
• langchain-classic
• nest_asyncio

API key:

• GOOGLE_API_KEY or GEMINI_API_KEY

Run:

python Code_8.py

Notes:

• this example uses simulated search results rather than live web search
• it still needs a Gemini API key because the agent itself is model-backed

Code_9.py

A CrewAI single-agent workflow example using OpenAI.

What it does:

• defines a stock-price lookup tool with simulated prices
• creates a financial analyst agent
• asks the agent to retrieve and report the simulated AAPL price
• checks for OPENAI_API_KEY before starting the crew

Key libraries:

• crewai

API key:

• OPENAI_API_KEY

Run:

python Code_9.py

Notes:

• CrewAI and the current Google ADK version require different OpenTelemetry versions
• if dependency resolution becomes unstable, use a separate virtual environment for this CrewAI example

Code_10.py

A Google ADK search-agent example using Gemini and Google Search.

What it does:

• creates a basic ADK agent with the built-in Google Search tool
• sends a science-news query to the agent
• prints the final response from the ADK runner

Key libraries:

• google-adk
• google-genai
• nest_asyncio

API key:

• GOOGLE_API_KEY or GEMINI_API_KEY

Run:

python Code_10.py

Notes:

• this example uses the Google Search tool through ADK
• it depends on Gemini API access, tool availability, and quota

Code_11.py

A Google ADK code-execution calculator example using Gemini.

What it does:

• creates an ADK LlmAgent with BuiltInCodeExecutor
• asks the model to write and execute Python code for math expressions
• prints event metadata, debug output, and final response text

Key libraries:

• google-adk
• google-genai
• nest_asyncio

API key:

• GOOGLE_API_KEY or GEMINI_API_KEY

Run:

python Code_11.py

Notes:

• this example uses built-in code execution through Google ADK
• it depends on Gemini API access and quota

Setup

Create and activate a virtual environment, then install the base dependencies:

python -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -r requirements.txt

Code_6.py also uses python-dotenv so it can load OPENAI_API_KEY from a local .env file.

Environment Variables

OpenAI examples

For Code_1.py, Code_2.py, Code_4.py, Code_6.py, and Code_9.py:

export OPENAI_API_KEY="your_openai_api_key"

Gemini / Google ADK examples

For Code_3.py, Code_5.py, Code_7.py, Code_8.py, Code_10.py, and Code_11.py:

export GOOGLE_API_KEY="your_google_api_key"

You can also use:

export GEMINI_API_KEY="your_google_api_key"

Current Dependency Status

The current requirements.txt in this repository contains:

langchain-core
langchain-openai
langchain-classic
langchain-google-genai
google-adk
google-genai
opentelemetry-api>=1.36.0,<1.39.0
opentelemetry-sdk>=1.36.0,<1.39.0
opentelemetry-exporter-otlp-proto-http>=1.36.0,<1.39.0
protobuf>=6.31.1,<7.0.0
nest_asyncio
python-dotenv

This means:

• the LangChain and ADK examples are represented in requirements.txt
• langchain-google-genai supports the Gemini-backed LangChain example in Code_8.py
• python-dotenv is included to support the .env loading used by Code_6.py
• OpenTelemetry packages are constrained to versions compatible with the current google-adk release
• Code_9.py uses CrewAI, which may need a separate environment because current CrewAI releases can require older OpenTelemetry packages than Google ADK

Summary

This repository currently demonstrates:

• sequential LangChain workflows
• manual routing with LangChain
• parallel LangChain processing
• reflection loops with LangChain
• coordinator-style agent orchestration with Google ADK
• parallel multi-agent research and synthesis with Google ADK
• sequential draft-and-review pipelines with Google ADK
• Gemini-backed LangChain tool calling
• CrewAI task orchestration with a custom tool
• Google ADK search and code-execution agents