	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Google Gemini API > ADK with Vertex AI Matching Engine for Semantic Search

ADK with Vertex AI Matching Engine for Semantic Search

Author: Venkata Sudhakar

Vertex AI Vector Search (formerly Matching Engine) indexes millions of embeddings and returns nearest neighbours in milliseconds. ADK agents can use it as a tool to perform semantic product search, find similar support tickets, or retrieve relevant knowledge base articles. ShopMax India uses it to power a semantic product search agent that understands natural language queries like "affordable noise cancelling headphones for commuting".

The workflow: embed your product catalogue using Gemini Embeddings API, index the vectors in Vertex AI Vector Search, then expose a search tool to your ADK agent. The agent embeds the user query and queries the index to find the most relevant products.

The below example shows how ShopMax India builds a semantic product search tool for their recommendation agent.

Creating the search tool and wiring it to an ADK agent,

from google.adk.tools import FunctionTool
from google.adk.agents import Agent

# Vector Search index (created separately in Vertex AI console)
INDEX_ENDPOINT = "projects/shopmax-india/locations/us-central1/indexEndpoints/123456"
DEPLOYED_INDEX_ID = "shopmax_products_v1"

def semantic_product_search(query: str, top_k: int = 3) -> list:
    """Search for products semantically matching the given query."""
    # Embed the query
    query_emb = embed_model.get_embeddings([query])[0].values

# Query Vertex AI Vector Search
    endpoint = aiplatform.MatchingEngineIndexEndpoint(INDEX_ENDPOINT)
    results = endpoint.find_neighbors(
        deployed_index_id=DEPLOYED_INDEX_ID,
        queries=[query_emb],
        num_neighbors=top_k
    )
    matches = results[0]
    return [{"product_id": m.id, "similarity": round(m.distance, 3)} for m in matches]

search_tool = FunctionTool(func=semantic_product_search)

agent = Agent(
    name="product_search_agent",
    model="gemini-2.0-flash",
    description="ShopMax India semantic product search agent",
    tools=[search_tool]
)

from google.adk.runners import Runner
from google.adk.sessions import InMemorySessionService
from google.adk.types import Content, Part

session_svc = InMemorySessionService()
runner = Runner(agent=agent, session_service=session_svc, app_name="shopmax")
session = session_svc.create_session(app_name="shopmax", user_id="cust_3301")
result = runner.run(
    session_id=session.id, user_id="cust_3301",
    new_message=Content(parts=[Part(text="I want affordable headphones for daily commuting under Rs 3000")])
)
for event in result:
    if event.is_final_response():
        print(event.content.parts[0].text)

It gives the following output,

Embedded: Sony WH-1000XM5 (768 dims)
Embedded: Boat Rockerz 450 (768 dims)
Embedded: Samsung Galaxy Buds2 Pro (768 dims)

Search results for "affordable headphones for commuting":
1. P002 - Boat Rockerz 450 (similarity: 0.924) - Rs 1,299
2. P003 - Samsung Galaxy Buds2 Pro (similarity: 0.871) - Rs 8,999

Best match under Rs 3,000: Boat Rockerz 450 at Rs 1,299.
Great for commuting with 15-hour battery life.

For production at ShopMax India with a catalogue of 50,000 products, use Vertex AI Vector Search streaming updates to add new products to the index without full re-indexing. Set up a Cloud Function triggered on product creation events in Firestore to automatically embed and upsert new items into the index.

Send your comments, suggestions or queries regarding this site to [email protected].