Product architecture

From QR to helpful, in five moves

ScanMate is a mobile-first assistant network for physical stores. The system is designed for zero-install access, per-store personality, and a tight latency budget so shoppers stay in flow.

Shopper flow

1. Scan

Shopper points camera at an in-store QR code.

2. Session

Web session boots on the shopper's phone — no install.

3. Context

Store, aisle, and product context load with the assistant.

4. Assist

AI answers with tone, catalog, and safety rules applied.

5. Insights

Every interaction feeds analytics and moderation.

System layers

Client (Mobile-first PWA)

QR scannerChat UIFavoritesStore finder

Edge API

Session brokerRate limitsSigned URLs

AI Engine

Persona + toneCatalog RAGGuardrailsFallbacks

Data & Analytics

ScansConversationsFeedbackCSAT

Rules engine

Persona lock

System prompt is per-store; overrides are ignored.

Catalog-first

Recommendations must reference in-stock items.

Latency budget

First token < 800ms, full reply < 2.5s.

Safety net

Sensitive queries route to human-approved answers.

Request lifecycle

Shopper phone ── QR ──▶ Edge API ── verify + session ──▶ AI Engine
                                    │                       │
                                    │◀── catalog + persona ─┤
                                    │                       ▼
                                    │                  Guardrails
                                    │                       │
                                    └──── stream reply ◀────┘
                                         │
                                         ▼
                                   Analytics + moderation

42ms

Avg. transport

780ms

First-token latency

99.98%

Rule engine uptime