Case Study · E-commerce AI
AI Product Assistant
An AI assistant built directly into e-commerce product pages.
This project was built to help online shoppers get answers about a specific product without leaving the product page.
The assistant uses the store's own product data and documentation to answer questions about specifications, variants, availability, materials and other product-related details. It was designed as a multi-tenant platform, allowing multiple stores to manage their own catalog data, widget configuration and assistant settings.
The Challenge
E-commerce product pages often contain a large amount of information, but customers may still have questions before making a purchase.
A general AI chatbot was not a suitable solution because answers needed to remain focused on the product currently being viewed. The platform also needed to support different stores without requiring every client to provide direct database access.
Another challenge was handling real-world product data, including product variants, XML feeds, incomplete data and additional product documentation.
The Solution
I worked on a product assistant that can be embedded directly into an e-commerce website.
The system identifies the current product page and retrieves information related specifically to that product. Depending on the type of question, the system can either return information directly from the indexed product data or use an AI model to generate a grounded response.
The platform also includes an administration area where stores can manage catalog sources, review conversations, configure quick questions and monitor indexing jobs.
Key Features
Focused answers, flexible onboarding.
Product-Specific AI Assistant
Answers questions about the product currently being viewed instead of searching across the entire catalog — keeping conversations relevant to the product page and reducing the risk of unrelated answers.
Flexible Catalog Integration
Product data can be indexed from either a database connection or a product feed — making it possible to onboard stores that were not comfortable providing direct database access.
Hybrid Product Search
The retrieval system combines semantic search with keyword-based matching, supporting both natural language questions and product-specific searches involving names, SKUs or technical terms.
Embeddable Product Widget
The assistant is added to a storefront with an embed script and displayed directly on product pages. Each tenant configures its own branding and widget styling.
Product Variants Support
Handles product variants and child SKUs, so the assistant works even when a shopper is viewing a specific variation of a parent product.
Admin & Monitoring Tools
The administration dashboard includes chat history, user roles, catalog settings and reindex job monitoring, with long-running indexing processes reporting their progress.
The Platform
Management tools for the business team.
Technical Challenges
Real-world constraints, practical solutions.
Keeping answers focused on the current product
The assistant resolves the product being viewed first and restricts the answer flow to information associated with that product. Factual questions such as availability or variant information can be handled directly from indexed data.
Supporting stores without database access
The platform supports product feed URLs as an alternative source — the ingestion pipeline processes the feed and prepares the product data for indexing and retrieval.
Working across different storefronts
Embedding into third-party sites introduced Content Security Policies, browser caching and inconsistent structured data. The widget works independently from the store's main frontend while detecting product information from the page.
My Role
Across the full stack.
I worked across the full application stack, from backend services and AI retrieval to the admin interface, embeddable widget and deployment.
- FastAPI backend services & APIs
- React administration dashboard
- Embeddable product assistant widget
- Product retrieval & indexing workflows
- OpenAI service integration
- PostgreSQL & Qdrant
- Magento product data ingestion
- Docker deployment & server configuration
- Production troubleshooting: storefront integration, caching, data quality
Technology
The stack.
Want an assistant on your product pages?
I build product-focused AI assistants that answer from your own catalog data — variants, feeds, and all.
Let's Talk