QA Manager → AI Systems

LLM Evaluation & Inference Performance Engineering

24 weeks at 15 hrs/week. Listed tasks total ~9–10 hrs most weeks — the other ~5 hrs are your overrun buffer, because hands-on GPU work always takes longer than planned. Check tasks off as you finish them; progress and your daily streak save locally and — once you sign in (top right) — sync across all your devices. Your next task is always pinned below.

Up next Loading…

Standing rules

Split the 15 hours deliberately. ~5 hrs on weekdays (1 hr/evening: reading, notes, small scripts) + two deep weekend blocks of ~5 hrs for hands-on work. Rent GPUs only during weekend blocks, when you can babysit the instance — never start one on a weeknight.
The buffer is for overruns, not new scope. If a week's tasks finish early, bank the time or start the next week — don’t add side quests. Falling one week behind is fine; falling three behind kills momentum.
Publish everything. Four honest write-ups with real numbers outrank any certificate. Each milestone below is a public artifact.
Never learn a concept without measuring it. Every optimization in months 3–4 gets validated with your own harness. That habit is what separates you from tutorial-followers.
Stop rented GPUs when you step away. The whole plan fits in $150–250 of compute only if instances don’t idle overnight.
Log what surprised you. One sentence per session in a running notes file. These become your blog posts and your interview stories.