Skip to content

About Agentic Search

An intelligent e-commerce discovery experience powered by Marqo's tensor search and AI summarization.

Agentic Search is a conversational search system that combines Marqo's tensor search with AI-powered summarization and query suggestions to create intelligent product discovery experiences. Unlike traditional search interfaces that rely on exact keyword matching, Agentic Search understands user intent and provides contextual guidance throughout the shopping journey.

Core Components

AI-Powered Summarization The system generates intelligent summaries that provide contextual product recommendations based on user queries. When a user asks "find me something to wear for a conference," the AI analyzes the query to understand the occasion, style requirements, and user preferences. It then generates a comprehensive summary that includes styling advice, product recommendations, and additional context like delivery options or availability. This goes beyond simple product listings to provide a personalized shopping assistant experience.

Smart Query Suggestions Agentic Search automatically generates two relevant follow-up queries to help users refine and explore their search. These suggestions are contextually aware and designed to guide users toward more specific searches or related product categories. For example, after searching for "conference attire," the system might suggest "women's formal dresses" and "men's suits" to help users narrow down their preferences. This feature reduces search friction and helps users discover products they might not have considered.

Tensor Search Integration The system leverages Marqo's advanced semantic search capabilities to find products based on meaning rather than exact keyword matches. This means users can search using natural language, synonyms, or even misspellings and still receive relevant results. The tensor search supports multimodal search across both text and images, allowing users to search for products using descriptive text while also understanding visual elements like color, style, or pattern.

Dynamic Categorization Search results are intelligently organized into relevant tabs and categories that adapt based on the query context. The system analyzes the user's intent and automatically creates meaningful groupings, such as "Women's Conference Attire" and "Men's Conference Attire" for professional clothing searches. This categorization maintains user context across different product categories and makes it easier to browse through relevant options.

Technical Architecture

The system processes natural language queries through a sophisticated multi-stage pipeline designed to maximize relevance and user experience:

Query Analysis extracts user intent, context, and specific product requirements from natural language input. This stage uses NLP techniques to understand not just what the user is looking for, but why they need it and what constraints might apply.

AI Processing generates contextual summaries and query suggestions using configurable AI models. This stage can be customized to use different language models based on your requirements, and includes prompt engineering to ensure responses are relevant to your specific product catalog and brand voice.

Tensor Search performs semantic search across your product catalog using Marqo's vector search capabilities. This stage finds products based on semantic similarity rather than keyword matching, ensuring that conceptually related products are surfaced even when users use different terminology.

Result Categorization intelligently organizes search results into relevant groups and tabs. This stage uses both the query context and product metadata to create meaningful categories that help users navigate through results more effectively.

Response Generation combines all the processed information into a structured response that includes the AI summary, query suggestions, and categorized product results. This final stage ensures that users receive a comprehensive, actionable response to their query.

Use Cases

Agentic Search is particularly effective in scenarios where users need guidance and context beyond simple product search:

E-commerce platforms benefit from enhanced product discovery that helps customers find exactly what they're looking for, even when they're not sure how to describe it. The AI summarization provides shopping guidance that can increase conversion rates and reduce support queries.

Fashion retail applications excel with style advice and occasion-based recommendations. Users can ask questions like "what to wear to a summer wedding" and receive both product suggestions and styling guidance tailored to the specific occasion.

B2B catalogs can provide intelligent product matching and suggestions for complex procurement needs. Users can describe their requirements in natural language and receive both product recommendations and technical guidance.

Content discovery applications can leverage multi-modal search across product catalogs to help users find items using both text descriptions and visual elements, making it easier to discover products through various search methods.

Agentic Search transforms traditional keyword-based search into an intelligent, conversational experience that helps users find exactly what they're looking for through natural language interaction, while providing the context and guidance they need to make informed decisions.

Quickstart

Follow these steps to get started with Agentic Search:

  1. Set up Marqo by running the Docker container and installing the Python client.
  2. Create a product index with appropriate settings for your catalog.
  3. Index your products with relevant metadata and tensor fields.
  4. Configure the agentic search with AI summary and suggestion settings.
  5. Test your implementation using the provided demo script.
  6. Integrate with your app using the API or custom implementation.

How Agentic Search works

Agentic Search brings together AI summarization, query suggestions, and Marqo's tensor search to create a flexible, production-ready conversational shopping experience.

Architecture overview

  1. User Input: The user sends a natural language query (e.g., "find me something to wear for a conference").
  2. AI Processing: The system generates contextual summaries and styling advice.
  3. Query Suggestions: The system suggests 2 relevant follow-up queries to refine search.
  4. Tensor Search: Marqo searches the product catalog using semantic similarity.
  5. Categorization: Results are organized into relevant tabs and categories.
  6. Response: The system returns AI summary, suggestions, and categorized product results.

Example flow

  1. User: asks "find me something to wear for a conference"
  2. AI Summary: Generates contextual advice about professional attire and available options.
  3. Query Suggestions: Creates follow-up queries like "women's formal dresses" and "men's suits".
  4. Marqo Search: Queries the product index for relevant items.
  5. Categorization: Organizes results into "Women's Conference Attire" and "Men's Conference Attire" tabs.
  6. Response: Returns the complete agentic shopping experience to the user.

Key concepts

AI Summary Contextual styling advice and product recommendations generated from user queries.
Query Suggestions 2 relevant follow-up queries to help users refine and explore their search.
Tensor Search Marqo's semantic search capabilities for finding relevant products.
Categorization Organization of results into relevant tabs and product categories.
Product Index Marqo index containing your product catalog with metadata and tensor fields.

What you can build

Agentic Search supports a wide range of conversational shopping experiences, including:

  • Shopping assistants that answer product questions and recommend items based on context.
  • Style advisors that provide fashion guidance and styling recommendations.
  • Conversational search interfaces for natural language product discovery.
  • Custom workflows using your own prompts, categorization rules, and integrations.

You can start from the provided demo template or build your own agentic shopping experience from scratch.

Next Steps

Ready to implement Agentic Search? Check out the Implementation guide to get started with setup, configuration, and API integration.