AI-Native Features

The ItPay PaaS Layer embeds three AI-native innovations that differentiate it from traditional payment gateways. These features leverage LLM capabilities — vision, natural language understanding, and intelligent routing — to reduce friction and enable new payment workflows.

1. QR Scanner AI

The QR Scanner AI enables users to pay by simply sending an image of a QR code to an AI agent. The system identifies the payment channel, extracts the amount and merchant details, and auto-routes the payment.

Flow Diagram

┌─────────────┐     ┌──────────────────┐     ┌──────────────────┐
│  User sends  │     │  LLM Vision      │     │  Channel         │
│  QR image    │────▶│  identifies:     │────▶│  Detection       │
│              │     │  • QR standard   │     │  Engine          │
│              │     │  • Channel       │     │                  │
│              │     │  • Amount (opt)  │     │  Match QR format │
│              │     │  • Merchant ID   │     │  to 13 channels  │
└─────────────┘     └──────────────────┘     └────────┬─────────┘
                                                       │
                                                       ▼
┌─────────────┐     ┌──────────────────┐     ┌──────────────────┐
│  User        │     │  PaymentIntent   │     │  QR Parsing &    │
│  Confirms    │◀────│  created with    │◀────│  Verification    │
│  & Pays      │     │  channel set     │     │  (CRC / sig)     │
└──────┬──────┘     └──────────────────┘     └──────────────────┘
       │
       ▼
┌─────────────┐
│  QR Charge  │
│  Generated  │
│  & Presented│
└─────────────┘

Scan Result Example

{
  "scan_id": "scan_9q3xr2m7kv1",
  "status": "recognized",
  "source": "image_upload",
  "raw_format": "raw_base64_encoded_jpeg",
  "detected_items": [
    {
      "channel": "alipay",
      "confidence": 0.987,
      "qr_type": "alipay_merchant_qr",
      "qr_version": "v2",
      "amount": 58.00,
      "currency": "CNY",
      "merchant": {
        "id": "2088421234567890",
        "name": "Coffee Express #12",
        "category": "food_and_beverage"
      },
      "payment_url": "https://qr.alipay.com/zkd9xm3k0e"
    }
  ],
  "alternative_channels_possible": [],
  "parsed_payload": {
    "vendor": "Alipay",
    "payload_format_indicator": "01",
    "point_of_initiation_method": "12",
    "merchant_account_info": "2088421234567890",
    "merchant_category_code": "5812",
    "transaction_currency": "156",
    "transaction_amount": "58.00",
    "country_code": "CN",
    "crc": "A32B"
  },
  "recommended_action": {
    "type": "create_payment_intent",
    "channel": "alipay",
    "amount": 58.00,
    "currency": "CNY",
    "merchant_id": "2088421234567890"
  }
}

Supported QR Formats

QR Format	Detection Accuracy	Processing Time
WeChat Pay QR	99.2%	< 800ms
Alipay QR (Merchant)	99.1%	< 800ms
Alipay QR (Personal)	98.7%	< 900ms
FPS HKQR	99.4%	< 700ms
PromptPay QR	99.0%	< 800ms
SGQR	98.5%	< 900ms
DuitNow QR	98.8%	< 850ms
QRIS	98.6%	< 900ms
VietQR	98.3%	< 950ms
GCash QR	97.9%	< 1s
InstaPay QR	97.8%	< 1s
Crypto Wallet QR	99.5%	< 500ms

2. NLP Payment

Natural Language Processing (NLP) payment enables users to instruct payments in plain language. The system parses the intent, extracts structured payment parameters, and executes the appropriate API call.

Supported Intent Types

Intent	Description	Example User Input
`pay`	One-time payment to a service	"Pay John's API service $5 for last month"
`subscribe`	Recurring subscription setup	"Subscribe to the premium tier for $15/month"
`query_invoice`	Check invoice or payment status	"What's my current balance on the translation service?"
`refund`	Issue a refund	"Refund the $30 overcharge from the image generation service"
`void`	Cancel an unpaid or pending payment	"Void the pending subscription for the email service"
`scan_qr`	Scan and pay via QR code image	"Scan this QR code and pay the 200 baht"

Intent-to-API Mapping

Intent	API Endpoint	Required Parameters	Response
`pay`	`POST /v1/payment-intents`	`service_id`, `amount`, `currency`	`PaymentIntent` with `status=requires_payment`
`subscribe`	`POST /v1/service-manifests/{id}/subscribe`	`service_id`, `plan_id`, `interval`	`Subscription` with `status=active`
`query_invoice`	`GET /v1/invoices?service_id={id}`	`service_id`, optionally `status`	Paginated `Invoice[]` list
`refund`	`POST /v1/payment-intents/{id}/refund`	`payment_intent_id`, `amount`	`Refund` with `status=completed`
`void`	`POST /v1/payment-intents/{id}/void`	`payment_intent_id`	`PaymentIntent` with `status=voided`
`scan_qr`	`POST /v1/ai/scan-qr`	base64-encoded image	`ScanResult` (see above)

NLP Processing Pipeline

User Input  ──▶  Intent Classifier  ──▶  Entity Extractor  ──▶  Parameter Builder
"Pay Alice       │                     │                       │
 $10 for         │ intent: "pay"       │ amount: 10.00         │ {
  the PDF         │ confidence: 0.96    │ currency: USD          │   service_id: "svc_..."
 tool"           │                     │ recipient: "Alice"    │   amount: 10.00,
                  └─────────────────────┘   item: "PDF tool"    │   currency: "USD"
                                                                 └─▶ POST /v1/payment-intents

Example Scenarios

Scenario 1: One-time payment

User: "Send $5 to the weather API for this month's usage"
→ Intent: pay
→ Entities: amount=$5, currency=USD, service="weather API", period="this month"
→ Action: Creates PaymentIntent for $5 USD routed to the weather API service

Scenario 2: Subscription

User: "Set up a weekly $2 subscription to the stock ticker service"
→ Intent: subscribe
→ Entities: amount=$2, currency=USD, interval=weekly, service="stock ticker service"
→ Action: Creates Subscription with weekly $2 billing cycle

Scenario 3: Multi-step

User: "Check my invoice and refund the last overcharge"
→ Step 1: query_invoice → finds last invoice with overcharge flag
→ Step 2: refund → issues refund for the flagged amount

3. Multi-Currency FX Routing

The FX Routing Engine intelligently selects the optimal payment channel and currency path when multiple channels are available for a given transaction.

Routing Decision Logic

                   ┌──────────────────────────┐
                   │  Payment Request          │
                   │  amount: 100 SGD          │
                   │  payee: Alice (HK)        │
                   │  payer: Bob (SG)          │
                   └────────────┬─────────────┘
                                │
                                ▼
                   ┌──────────────────────────┐
                   │  Channel Overlap Check   │
                   │                          │
                   │  Available Channels:     │
                   │  • PayNow (SGD→SGD)      │
                   │  • FPS (SGD→HKD via FX)  │
                   │  • USDC (SGD→USDC→HKD)   │
                   └────────────┬─────────────┘
                                │
                                ▼
                   ┌──────────────────────────┐
                   │  Fee Comparison           │
                   │                          │
                   │  PayNow: 0.0% + FX 0%     │
                   │  FPS: 0.25% + FX 0.8%     │
                   │  USDC: 0.3% + spread 0.1% │
                   └────────────┬─────────────┘
                                │
                                ▼
                   ┌──────────────────────────┐
                   │  Best Path:              │
                   │  PayNow (SGD→SGD)        │
                   │  Fee: 0%                 │
                   │  Route: payer → PayNow   │
                   └────────────┬─────────────┘
                                │
                                ▼
                   ┌──────────────────────────┐
                   │  QR Generated             │
                   │  SGQR with PayNow proxy   │
                   │  amount: 100 SGD          │
                   └──────────────────────────┘

Route Selection Criteria

Criteria	Weight	Description
Channel Overlap	Required	Both payer and payee must support at least one common channel
Fee	Primary (0.5)	Lowest total fee (channel fee + FX spread) wins
FX Rate	Secondary (0.3)	Best mid-market rate when FX is required
Settlement Speed	Tertiary (0.2)	Real-time settlement preferred over T+1
User Preference	Override	Explicit channel preference set by payer or payee

Settlement Mode Table

Mode	Payer Currency	Payee Currency	Typical Route	FX Required	Settlement Time
Local → Local	HKD	HKD	FPS (direct)	No	Real-time
Local → Local	SGD	SGD	PayNow (direct)	No	Real-time
Local → Local	THB	THB	PromptPay (direct)	No	Real-time
Local → Local	MYR	MYR	DuitNow QR	No	T+1
Local → Local	IDR	IDR	QRIS	No	T+1
Cross-Border	SGD	HKD	PayNow→ItPay FX→FPS	Yes (SGD→HKD)	Real-time
Cross-Border	THB	HKD	PromptPay→ItPay FX→FPS	Yes (THB→HKD)	Real-time
Cross-Border	CNY	HKD	Alipay→服务商 FX→FPS	Yes (CNY→HKD)	T+1
Cross-Border	CNY	SGD	Alipay→服务商 FX→PayNow	Yes (CNY→SGD)	T+1
Crypto	USDC	HKD	USDC→ItPay→FPS	No (stablecoin)	Real-time
Crypto	USDT	SGD	USDT→ItPay→PayNow	No (stablecoin)	Real-time
Crypto	USDC	USDC	On-chain transfer	No	Real-time (on-chain)
Crypto	USDT	HKD	USDT→ItPay FX→FPS	Yes (USDT→HKD)	Real-time

FX Rate Sources

Currency Pair	Primary Source	Secondary Source	Spread Target
HKD ↔ SGD	HKAB daily fixing	OANDA mid-rate	0.3% – 0.8%
HKD ↔ THB	BOT reference rate	Bloomberg	0.5% – 1.0%
HKD ↔ CNY	SAFE daily fixing	China Forex	0.2% – 0.6%
HKD ↔ MYR	BNM daily rate	Bloomberg	0.5% – 1.2%
HKD ↔ IDR	BI Jakarta Fix	Bloomberg	0.5% – 1.5%
HKD ↔ VND	SBV reference rate	Bloomberg	0.5% – 1.5%
HKD ↔ PHP	BSP reference rate	Bloomberg	0.5% – 1.5%
USDC ↔ Fiat	Coinbase / Binance spot	Kaiko	0.1% – 0.3%
USDT ↔ Fiat	Binance / OKX spot	Kaiko	0.1% – 0.3%
USDC ↔ USDT	DEX pool rate	Chainlink oracle	0.05% – 0.1%

Summary

These three AI-native features — QR Scanner AI, NLP Payment, and Multi-Currency FX Routing — form the core innovation layer of the ItPay PaaS. Together they enable:

Zero-UI payment initiation: send a QR image or type natural language
Intelligent channel selection: best price automatically chosen
Cross-border friction reduction: seamless multi-currency settlement

1. QR Scanner AI​

Flow Diagram​

Scan Result Example​

Supported QR Formats​

2. NLP Payment​

Supported Intent Types​

Intent-to-API Mapping​

NLP Processing Pipeline​

Example Scenarios​

3. Multi-Currency FX Routing​

Routing Decision Logic​

Route Selection Criteria​

Settlement Mode Table​

FX Rate Sources​

Summary​