Getting Started with LLMs

• Show the minimal code required to call an LLM.
• Demonstrate how to switch providers (OpenAI and Google Generative AI / Gemini) while keeping the same prompt and code flow.
• Highlight the environment variables and authentication patterns required for each provider.

​

1) Authentication and environment variables

# Bash examples
export OPENAI_API_KEY="sk-..."         # OpenAI
export GOOGLE_API_KEY="ya29...."       # Google Generative AI (API key)
# or
export GOOGLE_APPLICATION_CREDENTIALS="/path/to/service-account.json"  # for service accounts

• Google Generative AI docs: https://developers.generativeai.google/
• LangChain Python docs: https://python.langchain.com/en/latest/

import os
# os.environ["OPENAI_API_KEY"] = "sk-..."
# os.environ["GOOGLE_API_KEY"] = "ya29...."

​

2) Imports and minimal LangChain pattern

1. Import the provider wrapper (e.g., OpenAI or GoogleGenerativeAI).
2. Instantiate the wrapper (optionally specifying model and parameters).
3. Call the wrapper with a prompt and handle the response.

​

3) Example: OpenAI LLM (text-generation)

from langchain import OpenAI
import os

# Optional: set API key in-session (prefer exporting in your shell)
# Instantiate the OpenAI wrapper. You can pass model and parameters if needed.
llm = OpenAI()  # Uses OPENAI_API_KEY from environment; defaults to a model configured by the LangChain package
prompt = "What would be a good company name for a company that makes toys for kids?"

# Call the LLM and print the result
response = llm(prompt)
print(response)

Playful Pals Toys

• If you need deterministic output or other behavior, pass parameters like temperature or an explicit model when instantiating OpenAI(temperature=0.0, model="gpt-4o-mini").
• Some LangChain releases also expose chat-specific wrappers (e.g., ChatOpenAI) which are better suited for chat-native models.

​

4) Example: Google Generative AI (Gemini)

from langchain import GoogleGenerativeAI  # confirm exact import in your LangChain version
import os

# Optional: set in-session (prefer exporting in your shell)
# os.environ["GOOGLE_API_KEY"] = "ya29...."
# Instantiate with an explicit Gemini model
llm = GoogleGenerativeAI(model="gemini-pro")
prompt = "What would be a good company name for a company that makes toys for kids?"

response = llm(prompt)
print(response)

1. Joyful Creations
2. Imagination Unbound
3. Wonder & Play
4. Kiddie's Delight
5. Playful Wonders
6. Tiny Treasures
7. Learning Adventures
8. Whimsical Wonders
9. Spark of Imagination
10. Happy Hearts

• Choose the Gemini model appropriate for your use case (e.g., gemini-ultra, gemini-pro, etc.). Model names and availability can vary by account and LangChain version.
• If authentication fails, verify the environment variable method (API key vs service account) required by your client and LangChain integration.

​

5) Switching providers on the fly

# from langchain import OpenAI
from langchain import GoogleGenerativeAI
import os

# llm = OpenAI()
llm = GoogleGenerativeAI(model="gemini-pro")
prompt = "What would be a good company name for a company that makes toys for kids?"
print(llm(prompt))

​

6) Quick comparison: OpenAI vs Google Generative AI (Gemini)

​

7) Best practices and final notes

• Use environment variables for credentials; never commit keys to source control.
• Pass model-specific parameters (temperature, max tokens) at instantiation when deterministic or constrained behavior is required (e.g., OpenAI(temperature=0.2)).
• Prefer chat-specific wrappers for chat-native models when available.
• This lesson focuses on invoking LLMs with simple prompts. Subsequent lessons will cover chat-style APIs, embeddings, vector stores, and chaining LangChain components to build more advanced pipelines.

• LangChain docs: https://python.langchain.com/en/latest/
• OpenAI docs: https://platform.openai.com/docs
• Google Generative AI docs: https://developers.generativeai.google/