.svg)
We bring together the insights, talent, and technology to change the way the world travels. We're the engine and the engineers. We tackle the problems head on. In ways you've never seen.
You've just joined a small, high-ownership engineering team building a core product used by thousands of users every day. Over the last week, people keep saying "the product feels slow and flaky," but nobody owns this part of the system and there's no clear root cause. Metrics exist, but they're noisy, and no one has had time to dig in. Record a video walking us through how you'd approach this situation and what it says about how you work.
You're helping run operations for a product with a clear service-level expectation: most customer or internal tickets should be resolved within 2 hours. Over the last few weeks, the average resolution time has slipped to almost a full day, yet ticket volume hasn't obviously spiked. Some people blame "tool slowness," others say requests are "more complex now," but nobody has a clear, data-backed explanation. We're looking for real-world operational thinking, not theory --- assume this is happening in a small, fast, high-accountability team. Record a video walking us through how you'd diagnose this problem, what you'd do to stabilize it, and how your approach reflects how you like to work.
Scenario — "The Refund Mix-Up" Context: Axel's AI chat agent helps travelers understand their flight options including when they're eligible for a refund or credit. A customer sent this message: "Hi, my flight was canceled by the airline. Can I get my money back?" The AI responded: "Since your ticket is non-refundable, you are not eligible for a refund, but you can use the credit for future travel." The response was incorrect — when an airline cancels a flight, passengers are entitled to a refund, even on non-refundable tickets. Your Task: You've been asked to: Find what went wrong — what about the prompt or rule likely caused the incorrect reply? Fix it — write or describe an updated prompt snippet or instruction that would lead the model to give the right answer Test it — show what the correct AI reply should look like for the same user message Explain how you'd validate the fix before rolling it out (e.g., sample test chats, review process) Submission Guidelines Video Requirements: Your 7-10 minute video must include ALL three components: Introduction (1-2 minutes) Share one example of a time you improved how information was communicated — by a person, a system, or a process. What changed as a result? We want to know how you think. Traits Assessment (2-3 minutes) Which situation sounds more like the kind of work that energizes you — and why? Scenario A — "The Thursday Fix" (Execution): You notice the AI agent is giving wrong answers about baggage fees for one airline. You quickly verify the correct rule, update the prompt instructions, and confirm improved responses in testing. Would you enjoy that kind of direct, fast-turnaround problem solving? Scenario B — "The Pattern Builder" (Discovery): You see that refund-related mistakes are happening across multiple airlines. You spend time reviewing transcripts, identifying the root cause, and proposing a new set of logic or classification improvements. Would you prefer diagnosing larger patterns and designing fixes that scale? There's no right or wrong answer — just tell us which type of challenge fits how you work best. Challenge Response (3-5 minutes) Walk us through how you'd fix the faulty AI response in the scenario: What's wrong with the AI's answer? How would you adjust the prompt or rule? How would you test that it now works reliably? Keep it simple and practical — focus on your thought process and clarity. AI Usage Declaration: At the end of your video, please state: ☐ I did not use any AI tools ☐ I used AI tools (briefly describe how, including what I changed or adapted)
Description: Scenario — "Refund or Credit?": Context: Axel's AI travel agent helps users handle cancellations and rebookings automatically. Recently, users have received incorrect refund responses — for example: User: "My flight was canceled by the airline. Can I get my money back?" AI: "Sorry, your ticket is non-refundable, but you can use it as credit." That's wrong — airline-canceled flights should trigger refunds, even for non-refundable fares. Your Task: Design a lightweight Refund Decision State Machine or Refund API flow that helps the AI determine what to say reliably. Your system should: Accept key inputs (e.g., cancellation reason, fare type, airline policy data) Transition through simple states (e.g., checking policy → confirmed refund → safe fallback) Output a safe, correct response (refund / credit / escalate) Handle uncertain or missing data gracefully You don't need to build it — just design how it works and explain how you'd test it. Submission Guidelines Video Requirements: Your 7-10 minute video must include ALL three components: Introduction (1-2 minutes) Share one example of a time you automated something or made a system more reliable. What tradeoffs did you make, and what did you learn? Traits Assessment (2-3 minutes) Which situation sounds more like the kind of challenge you enjoy — and why? Scenario A — "The Friday Fix" (Execution): Customer support flags that refund messages are wrong for one airline. You're asked to fix the refund logic — one specific carrier, one intent — and deploy safely today. You'll test it end-to-end and verify nothing else breaks. Scenario B — "The Hidden Pattern" (Discovery): You notice refund messages vary depending on input phrasing. You propose a cleaner state machine that classifies each case — refundable, credit-only, unclear — before generating a response. You present it to the team as a small reliability improvement project. There's no right answer — just explain which fits your working style and why. Challenge Response (3-5 minutes) Walk us through your approach to the scenario: How would you structure the logic or API to handle refund eligibility safely? What states or conditions would it include? How would you handle "uncertain" or "missing data" cases? How would you test that it works before shipping? You can describe this as pseudocode, diagrams, or plain-language logic — we care more about clarity than syntax. AI Usage Declaration: At the end of your video, please state: ☐ I did not use any AI tools ☐ I used AI tools (briefly describe how, including what I changed or adapted)
