After a brief recruiter interview, I was given an open-ended work assignment with no recommended time limit. This involved two inference tasks, supervised and unsupervised, using what appeared to be production Typeform data. However, they added API development and deep learning models as "nice-to-haves." This was the first red flag—no respect for the interviewee's time at all.

I timeboxed my work to 4 hours per task, ignoring the "nice-to-haves" as I felt they went far beyond expectations for a work exercise. I was told the team was "very impressed" with my submission in any case.

I then underwent a two-hour interview with two technical staff members to go through the exercise. I received quite unprofessional comments during the interview, which gave me a bad feeling about the adequacy of the staff members to assess what I had done. For example, one interviewer claimed I should be using Logistic Regression to predict a continuous variable and stated there was no need to remove covariant features as it was a waste of time. Both were poor pieces of advice in the context of the task. They asked why I had not submitted a deep learning solution, and I explained that in inference tasks, I work up to more complex models from simpler ones, which is a common workflow, and that I had timeboxed the exercise.

The follow-up recruiter feedback was almost laughably unprofessional. I have a computing degree from a CS department that ranks at #10 worldwide, above Stanford, and 15 years of startup experience. The recruiter told me my "coding skills were lacking," that I did not complete the API assignment, and that I did not submit a deep learning solution, so they decided to reject me. This confirmed to me the completely amateur nature of the process. The job description did not even contain deep learning as a job requirement. By that stage, I had already decided I would not consider an offer in any case, given my interview experience, but that just rounded it off.

Typeform, please fix the alignment between your job descriptions, recruiters, interviewers, and work exercises for your ML roles! Candidates should not be expected to spare 3 days for unpaid work on production data.

I should add that I have been rejected from interviews only twice in my career, and it never left me feeling bad. I just wasn't suitable for those roles, and I felt the process was fair. Not so in this case; it was the worst interview experience of my life.

Why are you using an old model?

(I am not joking; I was asked this when reviewing interactive LDA output for unsupervised text classification.)