Exante’s software development engineer in test (SDET) Anatoly Bobunov has detailed how the financial services firm tore down its ageing test framework and rebuilt it from the ground up, designing the new architecture to scale alongside a growing team, an expanding suite of backend services, and the increasing role of AI coding tools.
The monolithic test repository at Exante had grown organically over several years, with different engineers introducing architectural changes at different times. Some decisions were carefully considered; others were quick fixes that quietly became permanent. Over time, responsibilities became blurred, layers interacted in unpredictable ways and business logic occasionally accessed the transport layer directly. Configuration was assembled in multiple places simultaneously, and the direction of dependencies grew increasingly unclear.
The consequences were tangible. Adding a new service required ever-more exceptions to the existing rules. Shared abstractions eroded, and changes to core components began affecting unrelated parts of the system. A particularly telling example was an attempt to integrate OpenTelemetry. What should have been a routine task became a high-risk operation because there was no single extension point. Several parallel branches of HTTP handling, each with different wrappers, meant the integration required touching many independent implementations across the codebase.
Exante also began moving towards auto-generated SDKs for testers, but plugging them into the existing framework would have required a partial redesign of the helper layer. At the same time, the team had begun using AI-powered coding tools, which struggled with the heterogeneous codebase. Without unified patterns or clear layer boundaries, AI tools had difficulty determining which practices were canonical — directly reducing the quality of generated output.
Rather than attempting incremental repairs inside the monorepo, Exante’s SDET lead Vladimir Smirnov gave Bobunov the time and space to design an entirely new framework in a separate repository. The strategy was to build a clean implementation and migrate automated tests across gradually, rather than halting development to carry out sweeping changes.
The new architecture is organised into strictly separated layers with directed dependencies flowing strictly top-down. At the domain level sit tests, helpers, a facade, endpoints and data models. Below that, a platform layer handles all infrastructure concerns, including HTTP transport, retries, timeouts, logging, tracing and error handling.
For more a detailed breakdown of the transformation, read the full story here.
Copyright © 2026 FinTech Global
