Opening your project
Real company data. Verifiable proof. On its way…
IBM Bob$20/monthFREE
Imagine it is your first week as a junior engineer at a German car-parts company, and an old program that decides when cars get serviced is quietly reporting wrong numbers. This is the time to show your skill. Get your own copy of the legacy repository, have IBM Bob read it, then fix and modernize the code with it, catching the mistakes it makes. Verify with the repo's own acceptance script and hand back your public GitHub link as proof.
Just completed
A graded ProoV project, completed step by step and marked against a published rubric.
Your first week at a car-parts company. Meet KM-Waechter, the small old service that decides when each car gets serviced, and the car it quietly let through.
Take the broken repository, spot the line producing the wrong wear percent, and explain in your own words why it reports 0 percent for a car that is nearly due.
Choose a prompt that bounds what the agent may change, then fix and modernize the whole repository with an AI coding agent and add a breakdown-risk analysis over the fleet history.
Review what the agent produced, catch the mistakes it planted, say why each one must not ship, and write the acceptance checklist the fix has to pass.
Run the repository's acceptance script, push to your own public GitHub, hand back the link, and explain what broke and what the AI got wrong in plain language a non-engineer could follow.
A car sits at 14,900 of its 15,000 km service interval and the service reports 0 percent worn. Whole-number division floors the wear, so the car is never flagged and it breaks down on the road. You find the line and explain why.
You scope the prompt, let the agent do the work, and hold the rules it must not touch: the 15,000 km interval and the 80 percent threshold stay exactly as they are.
You review a version of the agent's work and find what it actually did: it brought the rounding back, it moved a threshold nobody asked it to move, and it shipped no test proving the fix. You write the checklist it has to pass before you accept it.
A history of 120 labelled cars. Total mileage and age do not predict a breakdown; the gap since the last service and how hard the car is driven do. You rank the cars by risk and follow what the data says, not what everyone assumes.
The fixed code, the missing test, the risk analysis, and NOTES.md written in your own words about what the agent got wrong. You push it to your own public GitHub and hand back the link. The real code is what gets graded.
I gained real confidence in debugging and verifying code. The project taught me how to spot subtle mistakes, prove my fixes with tests, and explain the impact in plain language — skills that matter both for software quality and for teamwork.
I learned how to use an AI coding tool to understand and fix code. I also learned that I should always check the changes and run tests instead of blindly trusting the AI.
I learned how to use an AI coding agent effectively instead of blindly accepting its changes. I learned to review AI-generated code, identify bugs, check changes against the original requirements, and use tests to verify that the solution works correctly.
I learned how to use an AI coding agent effectively while still reviewing and testing its work myself. The project also helped me understand how to find silent bugs, verify fixes with tests, and use data to identify real risks instead of relying on assumptions.
Learned that "the AI said it's done" isn't proof of anything; I had to read every change Bob made and catch two things it got wrong on its own (a silent threshold change and a bug it claimed to fix but didn't). Also learned to spot a "silent" bug: one that never crashes, just qu…
Learned that "the AI said it's done" isn't proof of anything. I actually had to read every change Bob made and catch two things it got wrong on its own (a silent threshold change and a bug it claimed was fixed but wasn't). Also learned to spot a "silent" bug, one that never cras…
