Back to Home
Next.jsTypeScriptTailwind CSSGemma 4MCPDigitalOcean

Kapruka Conversational AI Agent

A conversational shopping agent that guides users from discovery to live payment gateway checkout entirely via chat. Integrated directly with Kapruka's Model Context Protocol (MCP) infrastructure, the system features a generative UI with visually rich animations, real time stock verification, and dynamic voice interaction.

Live Demo
kapruka-ai-agent.app / preview
Kapruka Conversational AI Agent

Key Features

  • Native multilingual support processing Sinhala, Singlish, English, and Tanglish.
  • Lightweight intent dispatcher to eliminate hallucinations before querying the LLM.
  • Real time delivery date and inventory validation against Kapruka's MCP endpoints.
  • Stateless cart memory persisting context via localStorage without costly session DBs.
  • Integrated browser compatible voice input (STT) and natural voice output (TTS).

Technical Challenge

The main challenge was handling mixed local languages (like Singlish) and out-of-stock items during a chat. I solved this by building a simple intent filter before calling the AI, and checking product availability directly with Kapruka's MCP.

"Focusing on system optimization and clean data patterns."