Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/Alejandrin08/Hackathon-SPEI/llms.txt

Use this file to discover all available pages before exploring further.

Before any transfer is confirmed, B-Accesible evaluates the transaction against a set of risk factors using an AI model. If the risk level is elevated, the user sees a plain-language explanation and is prompted to review before proceeding. Alerts are designed to inform, not alarm.

​
Risk scoring

Every transaction is evaluated by calling POST /ai/risk with context about the transaction and the user’s history. 
POST /ai/risk
Request body: 
{
  "amount": 15000.00,
  "is_new_beneficiary": true,
  "hour_of_day": 14,
  "num_past_transactions": 12,
  "avg_transaction_amount": 5000.00,
  "max_transaction_amount": 8000.00,
  "num_transactions_to_beneficiary": 0,
  "is_new_device": false,
  "geolocation_changed": false
}
Response: 
{
  "model_version": "risk-v1.0",
  "result": {
    "risk_level": "medium",
    "risk_score": 0.62,
    "risk_factors": ["NEW_BENEFICIARY", "AMOUNT_MUCH_HIGHER_THAN_AVERAGE"]
  }
}

​
Risk inputs

FieldTypeDescription
amountnumberTransaction amount in MXN
is_new_beneficiarybooleanWhether the recipient has not received a transfer from this user before
hour_of_dayinteger (0–23)Local hour when the transaction is initiated
num_past_transactionsintegerTotal number of past transactions by this user
avg_transaction_amountnumberHistorical average transaction amount for this user
max_transaction_amountnumberHistorical maximum transaction amount for this user
num_transactions_to_beneficiaryintegerNumber of prior transfers this user has made to this specific recipient
is_new_devicebooleanWhether the request originates from a device not previously seen for this account
geolocation_changedbooleanWhether the user’s detected location differs from their usual location

​
Risk levels

Risk score below thresholdThe transaction proceeds normally with no interruption to the user flow.No alert is shown. The RiskBadge component displays a green indicator.

​
Risk factor codes

The risk_factors array in the response contains one or more of the following codes. Each code corresponds to a specific condition detected by the model.
CodeCondition
AMOUNT_MUCH_HIGHER_THAN_AVERAGEThe transaction amount is more than 2× the user’s historical average
NEW_BENEFICIARYThis is the first transfer to this recipient from this account
UNUSUAL_TIMEThe transaction is initiated outside the 06:00–22:00 window
NEW_DEVICEThe request comes from a device not previously associated with this account
LOCATION_CHANGEDThe user’s detected geolocation differs from their typical location
LOW_HISTORYThe user has fewer than 3 past transactions, making pattern detection less reliable
Risk factor codes are used internally and in developer logs. The UI always shows plain-language descriptions, never raw codes.

​
UI components

  • RiskBadge — Displays the risk level visually (green / yellow / red) at the top of the confirmation screen.
  • ConfirmDialog — Presents the risk explanation and the list of contributing factors before the user confirms the transaction. Buttons are clearly labeled to avoid accidental confirmations.

​
Model performance

The risk model (risk-v1.0) was trained on synthetic transaction data.
MetricValue
Accuracy0.75
F1 score (macro avg)0.76

Build docs developers (and LLMs) love

Get started for freeTalk to us