AI Customer Support Agent with RAG Knowledge Base & Live System Integration

Built and deployed an AI-powered customer support chatbot for an e-commerce storefront. The system answers product questions using a curated knowledge base, retrieves real-time order information from WooCommerce, and escalates conversations to human support in HelpScout when needed. A built-in feedback loop helps surface knowledge gaps over time.

1. Knowledge-First Answers

All product and company-related questions are answered exclusively using a Pinecone-powered vector knowledge base, rather than relying on the LLM’s general training data. This approach helps ensure customers receive accurate, brand-approved information while significantly reducing the risk of hallucinated responses.

Curated Knowledge Base

The knowledge base is manually curated from official documentation, FAQs, and product descriptions. This keeps responses aligned with approved business information and avoids dependency on automatically ingested or unverified sources.

2. Live System Access

Customers can check their order status directly through the chatbot. The agent queries WooCommerce using the customer’s email address and returns a personalized summary of their recent orders.

Access to customer-specific data is protected, and order information is only returned when a matching customer record is found.

3. Smart Escalation

When the bot cannot confidently answer a question, it escalates the request to human support instead of guessing. In these cases, the system can:

• Create or check an existing support conversation in HelpScout
• Attach relevant conversation context and recent message history for support agents
• Log the unanswered question to a tracking sheet for later review

Self-Improving by Design

Every unanswered question is automatically logged and categorized by topic, such as Shipping, Product, Medical, Returns, or General. This log acts as an ongoing content gap report, making it easier to identify where documentation or FAQs should be expanded.

Session Handling

Conversation context is preserved using short-term conversational memory during the active session. No user authentication or long-term session binding is implemented, keeping the interaction frictionless while still maintaining continuity within a single conversation.

Business Impact

• Reduced HelpScout support ticket volume by approximately 40% by resolving common customer questions directly through the chatbot
• Improved first-response time for order status and product-related inquiries
• Gave the business clear visibility into what customers are asking, turning unanswered questions into actionable improvements for documentation and FAQs

Design Philosophy

The system is intentionally designed to prioritize correctness, traceability, and brand alignment over open-ended AI behavior. When confidence is low, the agent escalates to human support rather than attempting to infer an answer. This approach favors trust and customer experience over speculative responses.

Tech Stack

• Orchestration: n8n
• AI Models: OpenAI (GPT-4o and embeddings)
• Vector Database: Pinecone
• Integrations: WooCommerce REST API, HelpScout API, Google Sheets