
💹 Fintech & Capital Markets · Data Flow
How every payment is screened for fraud in real time using device data, velocity counters, rules and a model, with doubtful cases sent for review.
Drawing diagram…
Data flow for real-time fraud screening: a payment request with device fingerprint goes to the fraud service. It reads velocity counters (payments per card per hour) from Redis and the customer's history, runs a rules engine and a machine learning model, and combines the scores. Low risk is approved, high risk declined, and medium risk sent to a review queue for analysts. Decisions feed back as labels for model training.
flowchart LR P[Payment Request] -->|Amount, card, device| FS[Fraud Service] DV[(Device Fingerprints)] -->|Known devices| FS VC[(Velocity Counters - Redis)] -->|Payments per hour| FS CH[(Customer History)] -->|Past behaviour| FS FS -->|Features| RE[Rules Engine] FS -->|Features| ML[Fraud Model] RE -->|Rule score| DEC[Decision] ML -->|Model score| DEC DEC -->|Low risk| OK[Approve] DEC -->|High risk| NO[Decline] DEC -->|Medium risk| RQ[Review Queue] RQ -->|Analyst verdict| LBL[(Training Labels)]
A digital lending app that takes a loan application from KYC to disbursal in minutes, using credit bureau data, bank statement analysis and e-mandates for repayment.
How a lender gets a customer's bank statements through India's Account Aggregator network: consent request, approval in the AA app, and encrypted data sharing.
How a Buy Now Pay Later option decides at checkout whether to approve a shopper and splits the purchase into instalments.
The states of a stock market order in a broking app, from placement through exchange matching to partial and full fills.
Where a stock broking platform runs: low-latency order systems co-located near the exchange, and customer apps and back office in the cloud.
Tables for a loan book: borrowers, loan products, loans, repayment schedules, payments and collection actions.