Banks lose to fintechs because their software cycle is measured in quarters, not days. We bridge that gap for the banks we serve — core banking integration, AI risk models, and onboarding pipelines built to the same standards as their compliance teams demand, shipped at the pace their product teams need. Axis, Bajaj, IDFC First, HDFC, and Union Bank of India all run software we engineered.
Real-time transaction scoring, network analysis, mule detection. XGBoost ensembles tuned per channel. Sub-100ms scoring at 1.8M tx/day.
PD, LGD, EAD modeling for retail, MSME, and corporate. PSI/KS monitoring, vintage analysis, monthly recalibration cycles.
Aadhaar e-KYC, OCR pipelines, face liveness, dedup engines. 90-second onboards across 14 deployments.
Finacle, TCS BaNCS, Flexcube integrations. Account, transaction, GL, and limit APIs — wrapped, throttled, audited.
UPI, IMPS, NEFT, SWIFT MT & MX (ISO 20022), card switching. Reconciliation engines tested to T+0 close.
FIX 4.4/5.0 connectivity, OMS adapters, low-latency market data ingestion. Sub-millisecond paths where they matter.
Axis Bank's previous fraud engine fired on too many legitimate transactions and missed coordinated mule networks entirely. We rebuilt the scoring stack around an XGBoost ensemble with a graph-features layer for entity resolution. 94.2% recall at 0.4% false-positive rate, scoring at p99 < 80ms.
Every payment, every settlement, every ledger entry has a reconciliation harness that runs before code merges. If accounts don't balance to the rupee, the test fails.
Every monetary endpoint takes an idempotency key. Every retry is provably safe. We've never double-debited in production. Not once.
Every credit model ships with a calibration plot, a PSI dashboard, and a documented monthly recalibration trigger. The risk team owns the dashboard.