// loading
The track for the people building with LLMs and being honest about what works. Prompt engineering past one-shot tricks, model ops on a budget, evals that actually catch regressions, and the ethics conversation you can’t skip.
Each lesson is a 1-line briefing on what you'll practice. Read the briefing, then click Practice in terminal to drop into a real sandbox.
Prompt regression after model updates, hallucinations on real customer data, token-cost spikes. The work nobody told you was going to be 60% of the job.
Eval set design, tool-use streaming, PII redaction, open-weights vs hosted trade-offs. The boring infrastructure that lets your demo survive contact with users.
Prompt injection, vector DB drift at scale, end-to-end feature launch. The questions that decide whether your launch lands on the front page for the right or wrong reason.
The graded assessment uses the same terminal sandbox as these lessons : only it scores you on accuracy, methodology, tool fluency, communication, and real-world fit.