
🏦 Finance & Banking · Data Flow
How a bank scores every transaction for fraud in milliseconds: features from history, an ML model, rules, and a case queue for analysts.
Drawing diagram…
Data flow for real-time fraud detection: card and UPI transactions stream through Kafka to a scoring service, which pulls features (spend velocity, device, location, merchant risk) from a feature store, runs an ML model and business rules, and returns approve, decline or step-up authentication to the payment switch. High-risk cases go to an analyst case queue; analyst decisions are labelled and feed model retraining.
flowchart LR TX[Card and UPI transactions] -->|Events| K[Kafka] K --> SC[Scoring Service] FS[(Feature Store: velocity, device, location)] -->|Features| SC SC -->|Score| ML[ML Model] SC -->|Checks| RU[Rules Engine] SC -->|Approve / decline / step-up| SW[Payment Switch] SC -->|High-risk cases| CQ[Analyst Case Queue] CQ -->|Labelled decisions| LB[(Labelled Data)] LB -->|Retraining| TR[Model Training] TR -->|New model| ML K -->|History| FS
The steps behind a UPI payment: the payer's app, the payer's bank, NPCI's UPI switch and the payee's bank, from entering a UPI ID to the money arriving.
A payment gateway that lets online merchants accept cards and UPI: the checkout, the payment core with fraud checks, and the card networks and banks it connects to.
A personal loan from application to disbursal: KYC, credit bureau check, underwriting rules, manual review for borderline cases, and the final agreement.
What happens in the second after a card is tapped or entered online: the acquirer, card network and issuing bank authorise the payment, with 3-D Secure for online payments.
The main building blocks of a bank's core platform: accounts, deposits, loans, payments and the general ledger, used by branches, internet banking and mobile.
How a bank or fintech opens an account fully online: Aadhaar/PAN verification, video KYC, risk checks and account creation.