1/19
20 résumé-review questions mapped to Sofia’s recent Nextdoor, JSS research, and Nomad Futurist STAR stories. Answers are concise S/T/A/R/L prompts for spoken practice.
What problem were engineers facing before you built the Pipeline Tester?
S: Engineers used a backfill tool and often needed a merged PR to test new pipeline jobs. • T: Build a local way to test connected jobs before merge. • A: Built a Python/Airflow Pipeline Tester and a CLI that Claude could use from a Linear ticket, with status and log reporting in Slack. • R: Made connected tests locally runnable and reduced friction for users; no measured adoption claim. • L: A working tool also needs a usable path into engineers’ workflow.
What did you personally design and implement?
S: Engineers used a backfill tool and often needed a merged PR to test new pipeline jobs. • T: Build a local way to test connected jobs before merge. • A: Built a Python/Airflow Pipeline Tester and a CLI that Claude could use from a Linear ticket, with status and log reporting in Slack. • R: Made connected tests locally runnable and reduced friction for users; no measured adoption claim. • L: A working tool also needs a usable path into engineers’ workflow.
Walk me through the tool from a user’s selection to the test result.
S: The Pipeline Tester had disconnected start/end selections and recurring jobs that failed in a one-off DAG. • T: Reject bad pairs early and reproduce useful production behavior. • A: Filtered end-job choices to reachable downstream jobs; traced schedule mismatch and adjusted the temporary DAG and date/snapshot macro assumptions. • R: Invalid pairs were caught before DAG creation, and scheduling artifacts no longer caused false failures. • L: Validate cheaply before expensive work and separate job logic from scheduler context.
How did you determine which jobs belonged between a selected start and end job?
S: The Pipeline Tester had disconnected start/end selections and recurring jobs that failed in a one-off DAG. • T: Reject bad pairs early and reproduce useful production behavior. • A: Filtered end-job choices to reachable downstream jobs; traced schedule mismatch and adjusted the temporary DAG and date/snapshot macro assumptions. • R: Invalid pairs were caught before DAG creation, and scheduling artifacts no longer caused false failures. • L: Validate cheaply before expensive work and separate job logic from scheduler context.
How did you prevent someone from selecting jobs that weren’t connected?
S: The Pipeline Tester had disconnected start/end selections and recurring jobs that failed in a one-off DAG. • T: Reject bad pairs early and reproduce useful production behavior. • A: Filtered end-job choices to reachable downstream jobs; traced schedule mismatch and adjusted the temporary DAG and date/snapshot macro assumptions. • R: Invalid pairs were caught before DAG creation, and scheduling artifacts no longer caused false failures. • L: Validate cheaply before expensive work and separate job logic from scheduler context.
What was the hardest bug or unexpected behavior you encountered?
S: The Pipeline Tester had disconnected start/end selections and recurring jobs that failed in a one-off DAG. • T: Reject bad pairs early and reproduce useful production behavior. • A: Filtered end-job choices to reachable downstream jobs; traced schedule mismatch and adjusted the temporary DAG and date/snapshot macro assumptions. • R: Invalid pairs were caught before DAG creation, and scheduling artifacts no longer caused false failures. • L: Validate cheaply before expensive work and separate job logic from scheduler context.
Why did recurring jobs fail in a one-off test DAG, and how did you fix that?
S: The Pipeline Tester had disconnected start/end selections and recurring jobs that failed in a one-off DAG. • T: Reject bad pairs early and reproduce useful production behavior. • A: Filtered end-job choices to reachable downstream jobs; traced schedule mismatch and adjusted the temporary DAG and date/snapshot macro assumptions. • R: Invalid pairs were caught before DAG creation, and scheduling artifacts no longer caused false failures. • L: Validate cheaply before expensive work and separate job logic from scheduler context.
Why did you move logic out of the Airflow UI?
S: Engineers used a backfill tool and often needed a merged PR to test new pipeline jobs. • T: Build a local way to test connected jobs before merge. • A: Built a Python/Airflow Pipeline Tester and a CLI that Claude could use from a Linear ticket, with status and log reporting in Slack. • R: Made connected tests locally runnable and reduced friction for users; no measured adoption claim. • L: A working tool also needs a usable path into engineers’ workflow.
How did the Claude and Linear workflow actually execute a test?
S: Engineers used a backfill tool and often needed a merged PR to test new pipeline jobs. • T: Build a local way to test connected jobs before merge. • A: Built a Python/Airflow Pipeline Tester and a CLI that Claude could use from a Linear ticket, with status and log reporting in Slack. • R: Made connected tests locally runnable and reduced friction for users; no measured adoption claim. • L: A working tool also needs a usable path into engineers’ workflow.
How did you report progress, failures, and logs to users?
S: Engineers used a backfill tool and often needed a merged PR to test new pipeline jobs. • T: Build a local way to test connected jobs before merge. • A: Built a Python/Airflow Pipeline Tester and a CLI that Claude could use from a Linear ticket, with status and log reporting in Slack. • R: Made connected tests locally runnable and reduced friction for users; no measured adoption claim. • L: A working tool also needs a usable path into engineers’ workflow.
How did you encourage engineers to adopt the tool?
S: Engineers used a backfill tool and often needed a merged PR to test new pipeline jobs. • T: Build a local way to test connected jobs before merge. • A: Built a Python/Airflow Pipeline Tester and a CLI that Claude could use from a Linear ticket, with status and log reporting in Slack. • R: Made connected tests locally runnable and reduced friction for users; no measured adoption claim. • L: A working tool also needs a usable path into engineers’ workflow.
How did you choose which papers to include in the review?
S: In a small research team reviewing LLM refactoring, a teammate favored including BERT-style code-smell classification papers. • T: Agree on a defensible scope. • A: Listened to the case for inclusion, discussed evidence over two weeks, and documented why classification differed from LLM-performed refactoring. • R: The team excluded those papers and published the review in Journal of Systems and Software. • L: Respectful disagreement and explicit criteria strengthen team work.
Tell me about your disagreement over the BERT-style papers.
S: In a small research team reviewing LLM refactoring, a teammate favored including BERT-style code-smell classification papers. • T: Agree on a defensible scope. • A: Listened to the case for inclusion, discussed evidence over two weeks, and documented why classification differed from LLM-performed refactoring. • R: The team excluded those papers and published the review in Journal of Systems and Software. • L: Respectful disagreement and explicit criteria strengthen team work.
How did you assess impressive accuracy claims when sample sizes were small?
S: Some refactoring papers reported 100% or 96.8% results from narrow or small samples. • T: Interpret those numbers responsibly in the literature review. • A: Checked metric definitions, sample sizes, constraints, and leakage reporting; worked with coauthor to qualify claims, noting about 14 of 50 papers described formal leakage precautions. • R: The review presented high scores with their conditions rather than ranking them alone. • L: Verify what a headline metric actually measures.
What did you learn about data leakage or the limits of published results?
S: Some refactoring papers reported 100% or 96.8% results from narrow or small samples. • T: Interpret those numbers responsibly in the literature review. • A: Checked metric definitions, sample sizes, constraints, and leakage reporting; worked with coauthor to qualify claims, noting about 14 of 50 papers described formal leakage precautions. • R: The review presented high scores with their conditions rather than ranking them alone. • L: Verify what a headline metric actually measures.
What was your individual contribution to the published paper?
S: Some refactoring papers reported 100% or 96.8% results from narrow or small samples. • T: Interpret those numbers responsibly in the literature review. • A: Checked metric definitions, sample sizes, constraints, and leakage reporting; worked with coauthor to qualify claims, noting about 14 of 50 papers described formal leakage precautions. • R: The review presented high scores with their conditions rather than ranking them alone. • L: Verify what a headline metric actually measures.
Which project on your résumé are you proudest of, and why?
S: Engineers used a backfill tool and often needed a merged PR to test new pipeline jobs. • T: Build a local way to test connected jobs before merge. • A: Built a Python/Airflow Pipeline Tester and a CLI that Claude could use from a Linear ticket, with status and log reporting in Slack. • R: Made connected tests locally runnable and reduced friction for users; no measured adoption claim. • L: A working tool also needs a usable path into engineers’ workflow.
Tell me about a technical decision you changed after getting new evidence.
S: The Pipeline Tester had disconnected start/end selections and recurring jobs that failed in a one-off DAG. • T: Reject bad pairs early and reproduce useful production behavior. • A: Filtered end-job choices to reachable downstream jobs; traced schedule mismatch and adjusted the temporary DAG and date/snapshot macro assumptions. • R: Invalid pairs were caught before DAG creation, and scheduling artifacts no longer caused false failures. • L: Validate cheaply before expensive work and separate job logic from scheduler context.
Describe a time you disagreed with a teammate.
S: In a small research team reviewing LLM refactoring, a teammate favored including BERT-style code-smell classification papers. • T: Agree on a defensible scope. • A: Listened to the case for inclusion, discussed evidence over two weeks, and documented why classification differed from LLM-performed refactoring. • R: The team excluded those papers and published the review in Journal of Systems and Software. • L: Respectful disagreement and explicit criteria strengthen team work.
Tell me about a time you raised a concern others had missed?
S: A Nomad Futurist mockup used a brown accent as the dominant screen color, unlike the nonprofit’s website. • T: Raise the concern before implementation. • A: Showed the live site to the lead, suggested navy/gray/white, and offered to speak with design; lead relayed feedback. • R: Client also disliked the mockup; revised design was pending when this story was recorded. • L: Bring evidence and a user perspective when challenging a design.