NovaMind — AI Customer Assistant

## Overview NovaMind's support team was drowning in repetitive tickets. We built a RAG-based AI assistant trained on their documentation, past tickets, and product knowledge base. The bot handles tier-1 support autonomously, escalates edge cases with full context, and learns from every human resolution. ## Architecture - **LLM**: Claude Sonnet via Anthropic API with prompt caching - **RAG pipeline**: Pinecone vector store + custom chunking strategy - **Integrations**: Intercom (primary), Slack (internal alerts), Jira (auto-ticket creation) - **Deployment**: Serverless on AWS Lambda with a Redis cache layer ## Results - 74 % of tickets resolved without human intervention - Average first-response time: 8 seconds - Support team headcount frozen despite 3× user growth