All work

Automated and load testing of an AI chatbot

Automated and load testing for a public institution’s AI chatbot, run in parallel against production and beta and comparable question by question.

The challenge

Verifying the chatbot before a release and after it.

A public institution runs an AI chatbot answering questions from incoming foreign professionals, and needed to verify it before and after releases: that the answers stay correct across environments and that it holds up under concurrent use.

Our approach

A real browser for correctness, threads for load.

We built a Robot Framework suite driving a real browser: dismissing the cookie bar, asking a question, checking that the question rendered correctly, waiting for the answer and capturing it. The suite runs in parallel against production and the beta environment. Alongside it, a load script in Python.

The load script sends questions from a CSV across ten concurrent threads
Each question records both the answer and its measured time
A JMeter test plan
Answers from every run land in CSV, so production and beta compare question by question
Evidenced outcome

The repository holds the actual output of three runs: answers paired with a measured time for every question, roughly 223 rows for the performance run. The repository states no summary figure.

More work

All work →

Have a system that needs to just work?

hello@continero.com