FinTech

NullStack’s digital lending platform architecture implements RBI digital lending guideline requirements as application-layer business rules, not documentation policies. The loan origination workflow enforces the generation and borrower acknowledgement of a Key Fact Statement (KFS) containing APR, processing fees, and penal charges before the loan execution step can proceed. Disbursement routing is configured to flow from the Regulated Entity’s escrow account directly to the borrower’s verified bank account, bypassing the Lending Service Provider’s accounts entirely. Borrower communication templates — welcome messages, EMI reminders, overdue notices, and foreclosure statements — are maintained in a version-controlled template library with approval workflow, ensuring that no communication goes out in a format that has not cleared the compliance team. The audit trail covers every state transition in the loan lifecycle, every communication dispatched, and every document delivered, providing the evidence base that a regulatory inspection requires.

NullStack’s KYC onboarding flow integrates with UIDAI’s Aadhaar eKYC API for identity verification, NSDL and UTI for PAN verification, and a video KYC module (using the Daily.co WebRTC API) for product categories requiring in-person IPV. The onboarding UI is designed to minimise drop-off: each step collects only the information required for that step, progress is saved at each stage so a returning user does not restart from the beginning, and error messages are specific and actionable rather than generic. Fraud detection at onboarding uses a device fingerprinting layer and a velocity check against the phone number and Aadhaar hash to flag applications from devices or identities that have been seen in prior fraud events. Ongoing transaction monitoring uses a rule-based AML engine with configurable threshold rules aligned to the client’s RBI reporting obligations.

NullStack implements UPI payment collection via the Razorpay or Cashfree payment gateway, using server-side webhook processing with idempotency keys to handle the case where a webhook is delivered multiple times (a documented behaviour of payment gateway systems under high load). Every payment event — initiated, successful, failed, refunded — is written to an immutable ledger table in PostgreSQL with the gateway’s transaction reference and a timestamp, providing the complete record required for daily reconciliation against the gateway’s settlement report. Reconciliation is automated: a Python script runs nightly, matches application ledger records against the gateway settlement file, and flags any discrepancies — double credits, missing settlements, or refund failures — for manual review by the operations team. The reconciliation pass rate for NullStack-built payment systems is consistently above 99.8 percent before manual intervention.

NullStack’s alternative credit scoring pipeline is built as a Python microservice that ingests bank account statement data (via Account Aggregator framework integration, using NBFC-licensed AA access), GST return data (via GST API for SME borrowers), and optional supplementary signals, and outputs a risk score and recommended credit limit. The pipeline normalises transaction data into derived features — income regularity, salary date consistency, EMI obligation evidence, discretionary spending ratio — that a gradient boosting model (XGBoost or LightGBM) uses to produce a default probability estimate. Models are trained on the client’s own loan performance data as it accumulates, with periodic retraining scheduled via an MLflow-managed pipeline. The system is designed to meet RBI’s model explainability expectations for credit decisions, with SHAP-based feature importance outputs available per decision for audit purposes.