What is this application?

This is a web application built with Streamlit that demonstrates the Google Gemini API's grounding feature, specifically using Search-as-a-Tool with Google Search. It allows users to enter queries and receive AI-generated answers that are enhanced by real-time search results, displaying both the main answer and the related grounding information.

What do I need to run this app?

You need Python 3.x installed, the pip package installer, a Google Gemini API Key, and the 'streamlit' and 'google-generativeai' Python libraries. You can install the libraries using: pip install streamlit google-generativeai

How do I provide my Gemini API Key?

You have two options: 1. Set the 'GEMINI_API_KEY' environment variable in your system before running the app (recommended). 2. If the environment variable is not set, the application will prompt you to enter the API key in the web interface when you perform your first search.

How do I run the application?

First, save the Python code to a file (e.g., 'app.py'). Then, open your terminal or command prompt, navigate to the directory containing the file, and run the command: 'streamlit run app.py'. Streamlit will provide a local URL (like http://localhost:8501) to access the application in your web browser.

What is 'Grounding' or 'Search-as-a-Tool' in this context?

Grounding, or Search-as-a-Tool, is a feature of the Gemini API that allows the AI model to consult external data sources, such as Google Search, while generating a response. This helps to improve the factuality, accuracy, and relevance of the answer by incorporating up-to-date information from the web.

Can I change the AI model used by the app?

Yes, the Gemini model used can be changed by modifying the 'model_id' variable within the Python script. The default is set to 'gemini-2.0-flash', but you can replace it with another compatible model ID from the Gemini family.

Factual & Google Search Grounded Content Generation

lekhakAI

2 days ago3 min read

At ALwrity, Researching a topic with AI and generating factual and hallucination content is of paramount importance. Alwrity content generation depends heavily on web researched data from Google and AI researches. We have integrated Google grounded search in ALwrity to further factual & Hallucination free content generation with AI.

Gemini Grounding Search is a part of ALwrity application that allows users to interact with the Gemini API and leverage its powerful "Grounding" functionality. This application enables multi-turn searches and multi-tool queries, combining the power of Google Search, Grounding with Google Search, and content generation tools to deliver more accurate, informative, and contextually relevant responses.

The core purpose of this tool is to facilitate complex workflows that involve:

Enhancing factuality and recency through Grounding
Retrieving online artifacts for deeper analysis
Using media (images, videos, etc.) for multimodal reasoning or generation tasks
Conducting coding and technical troubleshooting
Accessing region-specific information
Translating content and finding relevant websites

With this tool, users can enrich search results by embedding real-time, web-based knowledge into their interactions with AI models.

Features

Grounding with Google Search: Seamlessly combine AI-generated responses with web-based, real-time data to enhance accuracy and relevancy.
Multi-turn Searches: Enables advanced workflows involving iterative questioning and response refinement.
Multi-tool Queries: Execute sophisticated queries by integrating multiple tools like Google Search and code execution.
Text and Content Generation: Generate accurate, relevant, and context-aware content based on the user's input and web search results.

Grounding with Google Search

Gemini Grounding Search uses the Grounding tool to enhance AI-generated responses by integrating search results from the web. This improves the accuracy, freshness, and factuality of the generated content, especially when answering specific queries.

Google Search Integration:

Free Tier: 1,500 Google Search queries per day (That's Generous).
Paid Tier: $35 per 1,000 additional queries.

This functionality works across all available languages, allowing for multi-language support and region-specific search capabilities.

ALwrity Grounding with Google Search

Grounding with Google Search is integrated within the tool to enhance the quality and accuracy of generated responses. By retrieving real-time data from the web, the tool can deliver answers that are up-to-date and more contextually relevant.

How Grounding Works:

Google Search: Each search query is supplemented by Google Search results, enriching the response with the latest, real-world information.
Languages: Grounding with Google Search supports all available languages, making it ideal for multilingual queries.

Code Breakdown

The application makes use of the google.genai Python client to integrate with the Gemini API. Here's an overview of the code:

Initialize Gemini Client

First, the Gemini API client is initialized using the provided API key:

client = genai.Client(api_key=os.environ['GEMINI_API_KEY'])

Perform the Search

Upon entering the search query and clicking the "Search" button, the application generates a response using the Gemini API, which integrates Google Search as a tool:

google_search_tool = Tool(google_search=GoogleSearch())
response = client.models.generate_content(
    model=model_id,
    contents=search_query,
    config=GenerateContentConfig(
        tools=[google_search_tool],
        response_modalities=["TEXT"],
    ))

Display Results

Once the search is performed, the response is displayed on the app, including the generated content and related search results:

st.header("Search Results")
if response.candidates[0].content.parts:    st.markdown(response.candidates[0].content.parts[0].text.replace('\n', '<br>'), unsafe_allow_html=True)

Grounding Metadata

If available, additional metadata, such as related search results, is displayed under the "Related Searches" header:

if hasattr(response.candidates[0], 'grounding_metadata'):    	       rendered_content = response.candidates[0].grounding_metadata.search_entry_point.rendered_content
t.markdown(rendered_content, unsafe_allow_html=True)

Conclusion

The Gemini Grounding Search app enhances search and content generation by integrating Grounding and Google Search, allowing users to access accurate, up-to-date, and relevant information. This tool simplifies complex workflows and improves the quality of AI-generated content, offering a seamless experience for both technical and non-technical users.

ALwrity