The experience our standards are built on

Testing a system with no single right answer

• Testing a chatbot and backend LLM services
• LLM observability and call tracing (Langfuse)
• Acceptance criteria and answer-quality evaluation
• AI quality in a regulated organization

Conventional testing assumes the correct result is known in advance. With a chatbot and backend LLM services it isn't, so the first job is agreeing what actually counts as a good answer. I lead QA in that area: the team defines the acceptance criteria, and we trace model calls in Langfuse so a bad answer can be followed back to what produced it. With an LLM, 'it looked fine to me' is not a test result.

Proving the data survived the migration

• QA strategy for a data modernization and migration program
• ETL pipelines on AWS: S3, Glue, Redshift, PySpark
• SQL-based data validation and reconciliation, data-quality checks
• Working in restricted-access environments with elevated security requirements

A large-scale data modernization and migration program in an enterprise environment. Moving the data is not the hard part; proving that what lands on the other side is exactly what was there before is. I own the QA strategy for the program and build the ETL pipelines on AWS, with validation and reconciliation run in SQL.

Testing IoT and metering platforms

• Automated testing strategy in Python and Selenium
• UI, functional and regression testing of web applications
• API and integration testing with SoapUI
• IoT, energy management and meter reading platforms

IoT, energy management and metering platforms: central measurement data repositories and meter reading systems. A failing test settles nothing on its own there — the question is still whether the fault sits in the application or in the meter. I led the automated testing strategy for those systems, and some of it stayed manual, because not everything in a system like that fits into a script. I also mentored members of the QA team.

Data pipelines and tests for a document platform

• ETL pipelines: batch and near-real-time processing
• SQL and PL/SQL transformations, processing and validation in Python
• Relational databases and MongoDB, Redis as a caching layer
• API, integration and end-to-end tests (Python, Selenium)

An enterprise document management platform: ETL workflows for batch and near-real-time processing. I designed, built and maintained those pipelines, and on the quality side I covered the same system. You write a pipeline differently once you are the one who has to break it.