Where to Find Pre-built LangChain Demos in the Google Cloud Generative AI Repository

The GoogleCloudPlatform/generative-ai repository does not contain an LLM-demos directory, but it provides extensive pre-built LangChain demos organized across workshops, search, and retrieval-augmented generation directories.

The GoogleCloudPlatform/generative-ai repository serves as a comprehensive resource for developers building generative AI applications on Google Cloud. While you won't find a dedicated LLM-demos folder, the repository contains numerous pre-built LangChain demos scattered across specialized directories for AI agents, Vertex AI Search, and RAG implementations.

Why There Is No LLM-demos Directory

A recursive search of the repository confirms that a directory named LLM-demos does not exist. The maintainers organize LangChain examples by functional domain—such as agents, search, and enterprise integration—rather than collecting them in a single monolithic folder. This modular structure places demos in contextually appropriate paths, making it easier to locate relevant code for specific use cases.

Where Pre-built LangChain Demos Are Located

The repository distributes working LangChain examples across several key directories. Each location targets specific integration patterns with Google Cloud services.

AI Agents Workshop (workshops/ai-agents/)

The notebook workshops/ai-agents/ai_agents_for_engineers.ipynb demonstrates end-to-end LangChain usage for building AI agents. It imports LLMChain and ChatPromptTemplate from LangChain, then constructs a multi-step essay-writing pipeline that combines Gemini with external tools like Tavily search.

Retrieval-Augmented Generation (search/retrieval-augmented-generation/)

The search/retrieval-augmented-generation/examples/question_answering.ipynb notebook provides a complete RAG implementation. It uses LangChain's RetrievalQA chain with a VertexAISearchRetriever to pull documents from Vertex AI Search, then generates answers using a Gemini LLM.

Vertex AI Search Integration (search/vertexai-search-options/)

Located at search/vertexai-search-options/vertexai_search_options.ipynb, this demo focuses on configuring LangChain with Vertex AI Search. It shows how to install the required LangChain packages and wire a VertexAISearchRetriever into a LangChain chain for document retrieval.

Dual LLM Comparison Demo (search/retrieval-augmented-generation/rag_with_dual_llms/)

This advanced demo, found in search/retrieval-augmented-generation/rag_with_dual_llms/, includes a Streamlit application (src/vertex_rag_demo_dual_llms_with_judge.py) that initializes two separate LangChain LLM instances—one using Vertex AI's text-bison model and another using Gemini. It runs both models on the same retrieved context to compare answer quality side-by-side.

Open Source Model Integration (search/gemini-enterprise/)

The notebook search/gemini-enterprise/oss_model_with_gemini_enterprise.ipynb demonstrates wrapping an open-source model with LangChain and exposing it through Gemini Enterprise, showing how to integrate non-Google models into the Google Cloud ecosystem using LangChain abstractions.

Code Examples from the Repository

Below are executable code snippets extracted from the notebooks listed above. These require installing langchain, langchain-google-vertexai, langchain-google-genai, and langchain-google-community.

Basic LLMChain with Gemini

This example from workshops/ai-agents/ai_agents_for_engineers.ipynb shows how to create a simple chain for generating essay outlines:

from langchain import LLMChain
from langchain.prompts import ChatPromptTemplate
from langchain_google_genai import ChatGoogleGenerativeAI

# Initialize Gemini model

gemini_llm = ChatGoogleGenerativeAI(model="gemini-1.5-flash")

# Define prompt template

outline_template = ChatPromptTemplate.from_messages(
    [
        ("system", "You are a helpful assistant that writes essay outlines."),
        ("human", "Create an outline for an essay about {topic}."),
    ]
)

# Create and run chain

outline_chain = LLMChain(llm=gemini_llm, prompt=outline_template)
outline = outline_chain.run({"topic": "the impact of AI on education"})
print(outline)

RAG Pipeline with Vertex AI Search

This snippet from search/retrieval-augmented-generation/examples/question_answering.ipynb implements a complete retrieval-augmented generation flow:

from langchain.chains import RetrievalQA
from langchain_google_vertexai import VertexAI
from langchain_google_community import VertexAISearchRetriever

# Configure Vertex AI LLM

vertex_llm = VertexAI(model_name="text-bison@001")

# Set up retriever for Vertex AI Search

retriever = VertexAISearchRetriever(
    project_id="my-gcp-project",
    location="us-central1",
    engine_id="my-search-engine",
)

# Build RAG chain

qa_chain = RetrievalQA.from_chain_type(
    llm=vertex_llm,
    chain_type="stuff",
    retriever=retriever,
)

# Execute query

answer = qa_chain.run("What are the security best practices for Cloud Run?")
print(answer)

Dual LLM Comparison with Streamlit

This example from search/retrieval-augmented-generation/rag_with_dual_llms/src/vertex_rag_demo_dual_llms_with_judge.py demonstrates initializing two LangChain LLMs for side-by-side comparison:

import streamlit as st
from langchain_google_vertexai import VertexAI
from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_google_community import VertexAISearchRetriever
from langchain.chains import RetrievalQA

# Initialize two different LLMs

vertex_llm = VertexAI(model_name="text-bison@001")
gemini_llm = ChatGoogleGenerativeAI(model="gemini-1.5-pro")

# Shared retriever

retriever = VertexAISearchRetriever(
    project_id="my-gcp-project",
    location="global",
    engine_id="my-search-engine",
)

def answer_with(llm):
    chain = RetrievalQA.from_chain_type(
        llm=llm,
        chain_type="stuff",
        retriever=retriever,
    )
    return chain.run(st.session_state.query)

# Streamlit UI

st.title("Dual‑LLM RAG Comparison")
st.text_input("Enter your query:", key="query")

if st.session_state.query:
    col1, col2 = st.columns(2)
    
    with col1:
        st.subheader("Vertex AI LLM")
        st.write(answer_with(vertex_llm))
    
    with col2:
        st.subheader("Gemini LLM")
        st.write(answer_with(gemini_llm))

Summary

• The GoogleCloudPlatform/generative-ai repository does not contain an LLM-demos directory.
• Pre-built LangChain demos are distributed across functional directories including workshops/ai-agents/, search/retrieval-augmented-generation/, and search/gemini-enterprise/.
• Key implementations include AI agent pipelines using LLMChain, RAG chains with RetrievalQA, Vertex AI Search integration via VertexAISearchRetriever, and dual-LLM comparison tools.
• All demos integrate with Google Cloud services including Vertex AI, Gemini, and Vertex AI Search using the langchain-google-* package family.

Frequently Asked Questions

Does the Google Cloud generative AI repo have an LLM-demos folder?

No, the repository does not contain a directory named LLM-demos. A comprehensive search of the repository structure confirms this folder does not exist. Instead, LangChain examples are organized by functional area across directories like workshops/, search/, and gemini-enterprise/.

Where can I find LangChain examples for Vertex AI?

You can find LangChain examples for Vertex AI in several locations. The search/vertexai-search-options/vertexai_search_options.ipynb notebook demonstrates basic Vertex AI Search integration. For RAG applications, see search/retrieval-augmented-generation/examples/question_answering.ipynb. The workshops/ai-agents/ai_agents_for_engineers.ipynb notebook also contains Vertex AI LLMChain examples.

Are there any RAG demos using LangChain in the repository?

Yes, the repository contains multiple RAG demos using LangChain. The primary example is search/retrieval-augmented-generation/examples/question_answering.ipynb, which implements a full RAG pipeline using RetrievalQA chains with VertexAISearchRetriever. Additionally, the rag_with_dual_llms directory contains a Streamlit application that compares two different LLMs using the same RAG retrieval context.

How do I run the dual LLM comparison demo?

To run the dual LLM comparison demo, navigate to search/retrieval-augmented-generation/rag_with_dual_llms/ and execute the Streamlit application located at src/vertex_rag_demo_dual_llms_with_judge.py. You will need to install dependencies including streamlit, langchain-google-vertexai, langchain-google-genai, and langchain-google-community. The application initializes two LangChain LLM instances—one using Vertex AI's text-bison model and another using Gemini—and runs both against the same Vertex AI Search retrieval context for side-by-side comparison.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}