Arnavbits26/self-improving-ai-interviewer ? reverse-engineered prompt

Build me a prototype AI interviewer that can learn from past interview sessions and then do better on new ones. I want it to read historical interview data, figure out what went wrong, turn those mistakes into general lessons, and use those lessons to ask better follow up questions and judge answers more accurately. It should also be able to run a fixed benchmark before and after learning so I can see whether it actually improved, not just memorized old transcripts.

Please make it work end to end with a simple way to run the full pipeline and a dashboard where I can explore the historical sessions, the lessons it learned, the benchmark results, and side by side transcripts from before and after learning. If there is no API key available, it should still run using a transparent offline fallback so I can demo it locally. Use current docs online if you need to, and keep the whole thing easy to launch.