The challenge: a critical system constrained by legacy technology

The platform managed payroll operations distributed across Latin America. Its calculation core had been developed in Visual Basic, while its interfaces relied on ASP Classic. Users needed old Internet Explorer versions or modern browsers running in compatibility mode.

Over the years, the system had accumulated formulas, taxes, labor agreements and country-specific exceptions. That knowledge was essential, but it lived inside a monolithic architecture in which screens, batch jobs and business rules evolved together. A regulatory change could affect multiple components and require extensive manual verification.

Replacing the interface alone would not solve the problem. The program needed to preserve existing payroll behavior, separate calculation from the web experience and migrate without disrupting payroll closing dates that could not be postponed.

The solution: a complete, traceable payroll flow built for regional operations

The flow begins when a payroll analyst signs in from a modern browser and selects the company, country, period and payroll type. The Next.js application adapts available actions to the analyst's role. NestJS receives payroll inputs through files or APIs and validates employees, concepts, formats, periods and duplicates before calculation can begin.

Errors are reported per record, so the user does not need to repeat an entire import. After validation, the platform creates an immutable snapshot of all inputs. A payroll run can therefore be reconstructed even if master data, formulas or regulatory rules change later.

NestJS sends the job to a queue, and the Go engine loads the rule version associated with the country, company and effective date. Calculation is divided into independent units so employees can be processed in parallel. Every run has an idempotency key: retries cannot duplicate movements or generate the same result twice.

When processing finishes, the analyst receives a preview covering earnings, deductions, taxes, employer contributions and net pay. The platform compares results with the previous period and highlights unusual changes or concepts outside expected ranges. From any row, the analyst can inspect the input, formula, rule version and calculated value.

A manual adjustment requires a reason and is linked to the user, date and previous value. A reviewer then resolves exceptions and approves or rejects each change. Once all controls have passed, an authorized user closes the period and the results become locked.

After closing, the platform generates payslips and output files. The operator distributes results to employee portals, accounting systems, banks and regulatory repositories. Each delivery reports its status; if a destination is unavailable, the platform retries without recalculating payroll.

Platform architecture

Next.js is distributed through CloudFront and protected by AWS WAF. NestJS handles APIs, authorization, validation and process coordination. Go workers execute calculations independently and can scale during payroll closing peaks.

Amazon SQS separates long-running jobs from web interactions. If a user loses connectivity or closes the browser, processing continues in AWS and the application retrieves its state when the user returns. Validations required for the next step run synchronously, while calculations, documents and exports run in the background.

PostgreSQL stores transactional data, while S3 stores versioned files and documents. Tenant context is enforced at both the API and data layers. Encryption, managed secrets, backups and audit records protect sensitive payroll information.

Observability assigns a correlation identifier to every request. A payroll run can be followed across the interface, API, queue, engine and integrations, while alerts report errors, retries and processing-time anomalies.

The implementation: progressive modernization without a disruptive replacement

Migration was organized by capability and country. The team first documented concepts, formulas, dependencies and exceptions in the existing system. Anonymized historical payroll runs became a test library that processed the same data through both the old and new engines.

Results were compared concept by concept before each stage was enabled. New screens initially coexisted with legacy functions through a compatibility layer. Each flow was then transferred to the modern services while a tested fallback remained available during the first closing cycles.

Infrastructure was defined as code, and services gained automated testing and deployment pipelines. The resulting platform preserves accumulated payroll knowledge, removes the Internet Explorer dependency and allows rules, screens and integrations to evolve independently.