MagicalCX
ShippedEmpathy-first AI customer support — omnichannel inbox, memory, handoff, and EFRO™ upsells.
Founding / primary full-stack engineer
magicalcx_case_study.sh
// Problem
Most support bots feel robotic, forget past conversations, and force customers to repeat themselves — while teams juggle separate tools for every channel.
// What I built
As founding engineer, I built MagicalCX from scratch at Pure Intelligence AI Technologies. The platform delivers human-like AI agents with context and empathy (not scripted replies), memory across conversations, a unified inbox for web, email, WhatsApp, Instagram, and Messenger, seamless human handoff with full context, the EFRO™ engine for smart upsells/cross-sells without being pushy, and dashboards for reporting. I remain primary committer across Voice (VAPI/Twilio), messaging pipelines, CRM, and billing surfaces.
// Architecture
Next.js frontend with TanStack Query for server state. Firebase for auth and real-time data. Qdrant powers vector search for RAG-based conversations. Voice flows through VAPI with Twilio for PSTN routing. WhatsApp and messaging channels integrate via dedicated inbox and broadcast pipelines. Upstash Redis handles caching and rate limiting. Billing and plan management tie into the core CRM layer.
// Stack
// Outcomes
- ▸ Built MagicalCX from scratch as founding engineer
- ▸ Human-like AI agents with context, empathy, and conversational memory
- ▸ Unified omnichannel inbox: web, email, WhatsApp, Instagram, Messenger
- ▸ Seamless human handoff with full conversation context
- ▸ EFRO™ engine for smart upsells and cross-sells during support
- ▸ Dashboards and reporting for ops visibility