Interview report
AI application engineer · Hiring manager round
The question
What did your model improvement achieve, and how did you measure it?
Condensed translation of the sample interviewer’s question.
The answer · English translation
“On the vulnerability-detection project we labelled more than 60,000 samples and fine-tuned a large model with LoRA, cutting the static-analysis false-positive rate from 65% to 25%. We also combined AST and CFG graph structures for code auditing, with multi-task instruction tuning plus chain-of-thought, and reached 98% accuracy in internal testing.”
Original Chinese TTS answer · Listen, then compare it with the feedback.
What was missing
The number needs evidence.
The model flagged the absence of an evaluation-set definition, a baseline comparison and analysis of failure cases.
Summary of the sample feedback, not a new evaluation.
What to practise next
Explain how you tested it.
The report recommends naming the test-set size, benchmark setup and metrics, then explaining a failure and how you handled it.
Practise: “Our test set contained ___. Compared with ___, we measured ___. It still failed when ___.”
Practice scaffold based on the feedback. Fill it with your own facts.
Actual test-session ratings