switchpay

Smart payment routing infrastructure, built to maximize authorization rates across PSPs.

What it does

switchpay routes each transaction to the optimal Payment Service Provider based on observed performance. The routing engine scores PSPs in real time using a weighted combination of success rate (60%) and latency (40%) computed over recent transaction history. When data is insufficient, it falls back to geographic heuristics by country and currency.

Architecture

• Backend : FastAPI with idempotency handling, exponential backoff failover, and a SQLite store (WAL mode)
• Routing engine : dynamic scoring over a configurable history window (HISTORY_WINDOW = 200), with static geo-fallback at cold start
• PSP layer : simulated providers (Stripe, Adyen, Wise, Rapyd) with differentiated performance profiles: success rates from 82% to 97%, latency from 80ms to 800ms
• Dashboard : real-time KPIs, PSP distribution, latency histogram, and routing opportunity detection
• Frontend : React + Chakra UI, deployed on Render

Known limits

The current scoring assumes PSP performance is stationary over the history window — which real-world data would not guarantee. Handling non-stationarity (concept drift, seasonality) and quantifying estimation uncertainty are open problems that motivated further study.

Stack

Python · FastAPI · SQLite · React · Chakra UI · Recharts · Render