AI SHOPPING CONCIERGE & MCP TOOLS
Kapruka Genie
Jun 2026
AI shopping concierge for Kapruka: full-screen chat that searches products, manages cart, checkout, pay links, and order tracking via Kapruka MCP tools.
Technologies
Project details
Overview
Kapruka Genie (Aladdin) is an AI shopping concierge for Kapruka. A full-screen chat searches products, updates the cart, starts checkout, issues pay links, and tracks orders through Kapruka MCP tools. Onboarding is trilingual (English, Sinhala, Tamil). LLM routing is hybrid — Gemini Flash and Claude — with a gift finder, voice STT/TTS, and an admin panel.
Built for the Kapruka Agent Challenge (deadline 30 June 2026).
Problem
E-commerce search and checkout still assume the shopper already knows the catalogue. A concierge that can act (search, cart, pay, track) via tools is closer to how people actually shop for gifts.
What I built
- Full-screen conversational shopping UI
- MCP tool calls for catalogue, cart, checkout, pay links, and orders
- Hybrid LLM routing (Gemini Flash / Claude)
- Gift finder flow and voice input/output
- Trilingual onboarding and an admin panel
Impact
- End-to-end agent demo against a real Sri Lankan commerce brand’s tool surface
- Practical MCP + multi-model routing, not a single chatbot prompt
- Voice and localization included, matching local shoppers
Key features
- Chat that actually mutates cart and checkout state
- Gift finder
- Voice STT/TTS
- Admin operations panel
- English / Sinhala / Tamil onboarding
Technologies
Next.js, React, JavaScript, Zustand, Google Gemini, Claude, conversational AI / speech, Vercel, GitHub.
My role
Agent architecture, tool wiring, LLM routing, and the concierge UX for the challenge submission.