Portfolio project

AI Customer Support + RAG

A production-style support automation system that answers from approved documentation, retrieves customer data, and escalates cases the AI should not handle alone.

Try Knowledge Base View Test Questions

Problem

Support teams repeatedly answer the same questions.

The system is designed to reduce repetitive support work while preventing unsupported AI answers.

Knowledge questions

Retrieve relevant documentation and answer from approved sources.

Customer questions

Look up synthetic customer/order records when a question requires account data.

Exceptions

Escalate billing disputes, security concerns, unsupported questions, and complex technical issues.

Architecture

Planned automation flow.

Customer message
      ↓
Website / support form
      ↓
n8n webhook
      ↓
Intent classification
      ↓
 ┌───────────────┬────────────────┬─────────────────┐
 │ Knowledge     │ Customer data  │ Human escalation│
 │ question      │ / order        │ / exception     │
 ↓               ↓                ↓
Supabase         Supabase         Ticket / alert
pgvector         SQL lookup
 └───────────────┴───────────────┘
                 ↓
              Gemini
                 ↓
        Grounded response
                 ↓
          Customer + logs

Test cases

What the finished system should demonstrate.

Test	Expected behavior
“How much is Growth?”	Retrieve pricing documentation and answer.
“How do I connect Slack?”	Retrieve integration documentation and answer.
“Where is FA-1001?”	Query customer/order data and answer from the result.
“I was charged twice.”	Identify billing exception and escalate to human.
“Does it integrate with a system not documented?”	Do not guess; explain that the knowledge base does not confirm it.