Case Study · AI Products
AI Chatbot & Knowledge Assistant
Helping users find answers and products through natural conversation.
This project was built for e-commerce businesses that needed a better way for customers to search for products and ask questions.
Instead of relying only on traditional keyword search, the platform combines AI-powered chat and product search using the same retrieval system. The assistant can use information from product catalogs, website content, and PDF documents to provide more relevant responses.
The platform was designed to support multiple stores, with each business having its own data, settings, branding, and usage controls. Store administrators can manage their content sources, monitor conversations, review search activity, and reindex their data from a central dashboard.
The Challenge
Traditional e-commerce search works well when customers know exactly what they are looking for. It becomes less useful when users search using natural language or ask more detailed questions. For example:
“Show me affordable 22mm cedar cladding that is currently in stock.”
The system needed to understand the important parts of the request — product type, size, price expectations, and availability. Another challenge was keeping the assistant's knowledge up to date: product information could come from Magento, WooCommerce, website pages, or PDF documents, while each store needed to remain isolated from the others.
The Solution
I worked on a multi-tenant AI platform that combines an embeddable chat assistant with an AI-powered storefront search experience.
The system indexes product catalogs and additional knowledge sources into a dedicated search collection for each tenant. When a user searches or asks a question, the application processes the request, identifies relevant filters, and retrieves matching information before generating the response. Both the chat assistant and the storefront search use the same retrieval pipeline, helping provide consistent results across the customer experience.
Key Features
One retrieval system, two experiences.
AI Chat Assistant
Customers ask questions in natural language and receive answers based on the business's indexed products, website content, and documents — the system retrieves relevant information before generating a response rather than relying on general AI knowledge.
AI-Powered Product Search
A storefront search experience that returns product suggestions as users type. Natural language queries run through a retrieval pipeline supporting filters like product type, category, price, size, and availability.
Knowledge Base & Document Indexing
Works with multiple content sources — product catalogs, Magento and WooCommerce data, WordPress content, website pages, and PDF documents — so product information and documentation become part of the searchable knowledge base.
Multi-Tenant Platform
Multiple businesses run on a shared platform, each with separate catalog data, settings, branding, usage controls, and vector collections.
Search & Usage Analytics
Administrators review search trends, popular terms, zero-result searches, product clicks, chat usage, and customer feedback — surfacing what users look for and where the catalog has gaps.
Reindexing & Knowledge Updates
Catalogs and knowledge sources are reindexed through tracked jobs, with progress and index health visible in the admin dashboard for managing large data updates.
The Platform
A complete control center, not just a widget.
Technical Challenges
The hard parts worth solving.
Understanding natural language product searches
Converting natural language into useful product searches combines deterministic rules, language understanding, and hybrid search with fallback levels — handling specific requests while still returning useful results for broader queries.
Keeping knowledge up to date
Product information and documentation change frequently. Source adapters and reindexing jobs update catalogs, website content, and PDF knowledge, with progress and index health visible to administrators.
Supporting multiple businesses securely
Because the assistant embeds on external websites, the platform includes tenant-specific data separation, CORS controls, usage limits, authentication roles, IP and country blocking, and reCAPTCHA protection.
My Role
Across the full stack.
- FastAPI backend services & APIs
- AI retrieval & search pipeline
- Qdrant vector search
- React administration dashboard
- Embeddable AI chat widget
- Storefront search functionality
- Multi-tenant PostgreSQL design
- Magento & WooCommerce integration
- Website & PDF content ingestion
- Analytics & conversation tracking
- Reindex jobs & progress monitoring
- Docker deployment & infrastructure
Technology
The stack.
Under the hood the platform also uses SQLAlchemy, Alembic, hybrid dense/sparse vector search, Playwright, PDF extraction, JWT authentication, APScheduler, SMTP, reCAPTCHA, and GeoIP services.
Need an AI assistant for your business?
I build AI chatbots and product search experiences around your own catalog, content, and documents.
Let's Talk