ලිLiya

Judge guide

How Liya works

Liya is an AI shopping assistant for Sri Lanka that turns a human gift story into a live Kapruka cart, checkout link and tracking flow.

Unlike e-commerce sites that show products, Liya decides what to buy and completes checkout in one guided conversation.

Everything shown here is live from the same system powering the demo — not a separate slide deck.

2-minute checkout promiseHosted MCP onlyZero login
ලි
Liya

“Tell me the situation. I’ll handle the shopping.”

Intent → understanding → live search → ranked shelf → cart → checkout
If you trust me on one pick, I’ll choose the safest gift and get you to payment fast.

Interactive flow

From messy request to real checkout

Step 1 of 6

Intent

“I forgot my wife’s birthday. Kandy tomorrow. Rs. 5,000.”

Judge-visible result in the product UI
Conversation start: Messy human request
Conversation startCUSTOM IMAGE

Messy human request

I messed up. Wife is angry. Kandy tomorrow. Rs. 5,000.

apologywifeKandytomorrow
Liya asks less, understands more1/6

Proof of intelligence

Before vs after Liya

One messy sentence becomes a decision, shortlist, cart and checkout path.

Apology gift

Before: I forgot my wife’s birthday

After: Best repair gift + checkout path

Urgent birthday

Before: Birthday, tomorrow, not sure

After: Delivery-safe shortlist

Cheap but premium

Before: cheap but premium

After: Tradeoff resolved calmly

Interactive judge lab

Tap a scenario and watch Liya’s reasoning

This mini-console makes small but competition-visible features explicit: intent extraction, strategy selection, MCP query shaping, trust labels, and next-step nudges.

Shopper says: “I forgot my wife’s birthday. Kandy tomorrow. Rs. 5,000.”
Detected context
apologywifeKandytomorrowRs. 5,000
Strategy

Relationship repair → roses + chocolate + safer delivery

MCP query shaped by Liyaapology roses chocolate sorry card gift
Trust + next action

Safe-to-buy if city/date verified. Next: Choose one standout or compare top 3.

Checkout readiness

72%

Feature map

What judges should notice

Emotion AI

Apology, urgency, romance and celebration change the shopping strategy.

Sinhala / Tamil / Tanglish

Local-language signals shape Liya’s tone and shopping flow.

Live MCP search

Uses the hosted Kapruka MCP endpoint for real product discovery and checkout.

Memory ranking

Preferences like chocolate, roses or minimal style influence ranking.

Voice

Browser voice input and output for a more natural assistant feel.

Reorder

Recently viewed products become quick buy‑again choices.

Delivery-aware filtering

City, date and delivery risk shape product trust and checkout readiness.

Security & Privacy

Hosted MCP boundary, no secret keys in the browser, and only session-local shopping memory.

+ Accessibility+ Fast Checkout+ No Login Required+ Hosted MCP Only+ Performance+ Scalability+ Conversion signals

Local voice + language

Built for how Sri Lankans actually ask

Sinhala, Tamil, Singlish/Tanglish and browser voice are shown as part of the shopping flow — not buried as a settings feature.

Dictate

Tap Voice, speak, edit, send.

Detect

Scripts + local words adjust tone.

Reply

Warm local assistant style.

What makes it different

Liya is a decision assistant, not a browsing system.

A normal shop asks users to filter. Liya asks for the human situation, chooses a safe path, and moves gently toward checkout.

Liya starts with the occasion, not a search bar.
Liya recommends one safest pick instead of dumping a catalog.
Liya keeps the shelf, cart and conversation alive together.
Liya completes checkout with a Kapruka payment link — no login needed.

What Liya is not

Misclassification guardrail

Not a chatbot widget added beside the website.
Not a fake catalog or mock checkout flow.
Not a new MCP server or Kapruka backend fork.
Not a generic LLM wrapper — the frontend orchestrates a shopping journey.

Architecture truth

Why this architecture wins

Browser UI → Next.js frontend → isolated MCP client → hosted Kapruka MCP → Kapruka backend
  • • MCP is the commerce backbone.
  • • The frontend orchestrates intent, memory, ranking and checkout UX.
  • • No backend fork, no MCP server, no extra database.
  • • Deterministic fallback keeps demos safe if live search is slow.
  • • Real checkout links still come from Kapruka MCP.

Proof of resilience

Break Liya — Judge Mode

These buttons run predefined chaos prompts. Liya should stay calm, anchor the flow, and continue shopping.

Emoji chaos

😀😀😀

Cheap + premium conflict

cheap but premium today

API question mid-flow

what’s the API?

Rapid topic switch

birthday → anniversary

Ready for the 2-minute judge path?

Runs the apology/birthday/Kandy/payment-link story with live MCP search and fallback safety.