Testing an AI assistant, working at InPost and the new QA skill set
Episode 211 h 3 min
- AI
- Architecture
- Career
In this episode
Michalina Graczyk explains how to test an AI assistant whose responses are non-deterministic. Using InPost's shopping assistant as an example, we discuss evals, reference datasets, safety and the skills QA professionals will need.
Discussion topics
- How do you test an AI assistant that returns non-deterministic results?
- What are evals, golden sets, robustness, safety and guardrails?
- How can an assistant's illegal or dangerous actions be constrained?
- Which testing tasks could AI agents take over first?
- Which QA skills will be needed in the coming years?
Listen on
Guest
- Michalina GraczykEngineering Manager @ InPost
About the guest
Engineering Manager at InPost and Quality Assurance Engineer for AI-powered products. She worked on beta testing the AI shopping assistant in the InPost app.
