MarqTune
AI model · Multimodal model · Text → Text by Marqo
Commerce Superintelligence trained on your catalog and your shoppers.
MarqTune is a dedicated AI model layer that trains custom intelligence systems for enterprise retailers using their own catalog and shopper data. It uses Generalized Contrastive Learning to deliver 73-78% better relevance than generic models for search, browse, and recommendations.
Model specs
The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.
- Languages
- English, Spanish
Category: Machine Learning
- Training Method
- Generalized Contrastive Learning (GCL)
- Input Modalities
- Text, Image, Behavioral Signals
- Query Capacity
- 1B+ queries per day
- Relevance Improvement
- 73% to 78% over generic baselines
- Deployment Options
- API, Shopify, Adobe Commerce, Salesforce Commerce Cloud
- Target Market
- Mid-market to large enterprise retailers
- Trained on your real product catalog and shopper behavior
- Uses Generalized Contrastive Learning (GCL) for relevance
- 73% to 78% relevance improvement over generic baselines
- Resolves cold start for new products instantly upon indexing
- Processes over 1 billion search queries per day
Overview
MarqTune is Marqo's dedicated AI model layer that trains custom intelligence systems for enterprise retailers. It moves beyond generic LLMs by training a purpose-built model on a brand's specific product catalog and real shopper behavior signals like clicks and purchases. The model uses Generalized Contrastive Learning (GCL) to understand the relevance spectrum of products, enabling it to interpret complex intent and visual attributes without manual synonym tables. MarqTune powers the entire discovery suite, including search, browse, and recommendations, ensuring that even new products are discoverable the moment they go live. It is designed for mid-market and large enterprise retailers to replace legacy keyword-based infrastructure with commerce-specific superintelligence.
Understanding and Retrieval
- Unified Multimodal Input: The model processes text and images as a unified input to understand complex visual intent and text refinements.
- Semantic Intent Understanding: MarqTune interprets context and synonyms to map queries to product attributes without relying on manual synonym tables.
- Natural Language Support: The model supports long-tail, voice, and natural language searches to interpret how shoppers naturally speak.
- Instant Cold Start Resolution: MarqTune understands new products from the moment they are indexed based on visual attributes and relationships to existing inventory.
Model Training and Intelligence
- Catalog-Specific Training: MarqTune is trained on a retailer's specific product catalog and real shopper interaction data rather than generic datasets.
- GCL Training Architecture: The architecture uses Generalized Contrastive Learning to optimize the model based on real purchase signals and shopper behavior.
- Behavioral Learning: The model continuously updates rankings based on real-time onsite engagement signals like clicks, carts, and purchases.
- Enterprise Scale Performance: MarqTune provides deep visual and semantic understanding for over 1 billion search queries per day.
Benchmarks
- Relevance improvement: 73% to 78% over generic baselines
Use cases
- Retailers train custom models on their own catalogs to improve search relevance by up to 78% over generic baselines.
- Fashion brands use multimodal processing to allow shoppers to search using both images and text refinements simultaneously.
- Ecommerce teams eliminate the cold start problem by making new arrivals discoverable through visual attribute understanding the moment they go live.
- Enterprise marketplaces resolve over 1 billion queries per day with deep semantic understanding of complex product codes and model numbers.
Specifications
- Modalities
- Text, Image, Behavioral Signals → Search Results, Rankings, Recommendations
- Access
- API and direct platform integration for Shopify, Adobe Commerce, and Salesforce Commerce Cloud
- Fine-tuning
- Trained on a dedicated AI model on your real product catalog and shopper behavior
- Training Method
- Generalized Contrastive Learning (GCL)
- Input Modalities
- Text, Image, Behavioral Signals
- Query Capacity
- 1B+ queries per day
- Relevance Improvement
- 73% to 78% over generic baselines
- Deployment Options
- API, Shopify, Adobe Commerce, Salesforce Commerce Cloud
- Target Market
- Mid-market to large enterprise retailers
How it works
- Install the Pixel: The Marqo pixel is installed to capture shopper interaction data such as clicks, carts, and purchases.
- Model Training: MarqTune trains a dedicated AI model on your specific product catalog and the captured shopper behavior signals.
- Deploy and Go Live: The custom model is deployed via API or one-click integrations to power search, browse, and recommendations.
FAQ
How does it handle new products with no sales history?
MarqTune solves the cold start problem by understanding new products' visual attributes and relationships to existing inventory the moment they are indexed, without needing historical clicks.
Who is MarqTune designed for?
MarqTune is designed for mid-market to large enterprise retailers and is often used to replace legacy tools like Solr, Elasticsearch, or Searchspring.
Added by AntSpace from public sources (marqo.ai).