Payment Optimization Engine

A global digital-payments company engaged Taller to build the authorization stack primitives and embed engineering capacity.

Every wrongly declined payment was a lost sale and a frustrated customer. At the company’s scale, across trillions in payment volume, even small rates of avoidable failures (a card flagged by mistake, stale credentials, a soft decline never retried) added up to enormous sums left on the table. The company’s system for maximizing successful payments, keeping stored card details fresh, predicting and preventing declines, retrying failed payments intelligently, and routing transactions along the best path, depended on several core components and ongoing engineering capacity in the credit-platform and machine-learning-routing teams. Taller was engaged to supply both.

Taller delivered and integrated five foundational building blocks across the authorization stack, then embedded engineers to keep the engine running. These included the system of record on Apache Fineract (the ledger that tracks balance, credit, and buy-now-pay-later) and the gRPC communication layer whose low latency made smart retries economical to run at scale. Taller also delivered the post-purchase notifications component that fed the company’s AI-powered cashback and smart-receipts products, migrating the legacy C++ code to Java in the process. On the payment-processing side, Taller stabilized the Ruby on Rails platform, refactored the legacy code, and led the migration to AWS with automated deployment, the processor surface where the optimizations actually took effect. Through the unified onboarding platform, Taller integrated the network-tokenization and account-updater connections that kept stored card credentials fresh. Beyond these builds, Taller engineers contributed directly to the company’s credit-platform Java team and to the machine-learning engineering behind its predictive-decline and smart-routing models.