An AI engineer wires up models that already exist. A machine learning engineer builds the model. Which one you hire decides whether you ship in two weeks or spend six months in a sandbox, and here is how I tell the two jobs apart.
The RTX 30-series has no hardware FP8, so I packed four FP8 values into one FP32 container and unpacked them inside a Triton kernel. On a client’s 3080s that cut the memory bottleneck and ran about 3.3x faster.
A price tracker taught me that a hand-built Google scraper dies within two days. Here is why I hand that job to SerpApi now, plus the PHP wrapper I reuse and the latency setting that matters at scale.
Seven avatars from one SVG symbol, each with its own animation. The CSS cascade dies at the Shadow DOM, but custom properties pass straight through it, which is how one symbol ends up blinking and tapping at a different rate per instance.
A client dropped 8MB 4K PNGs into the media library for holiday backgrounds and the LCP went red within the hour. Here is what fixed it: a 200KB cap on hero images, one file per breakpoint, and WebP.
A fleet dashboard flashed so hard that a broken-down truck went unnoticed for twenty minutes. Faster polling made it worse. What helped: delta indicators, sparklines, skeleton screens, an honest last-sync label and ARIA live regions.
Trimming a legacy site cut page weights by 80%, and the same work cut its energy use. Five steps: pick one journey, benchmark it, set a page weight target, strip the WordPress junk nobody needs, then keep checking.
A 10,000 SKU WooCommerce import kept hanging, and the previous dev’s answer was a 2GB memory limit. Advent of Code puzzles point at the real fix: sketch the logic first, validate the input, batch the writes.
A client paid thousands for an AI accessibility overlay and still got complaints from deaf users. Notes on what does help: text-based contact options, human captions, visual and haptic cues in place of sound, and accessibility handled before launch instead of after.
A client thought their WooCommerce recommendation engine was broken because accuracy at a 0.5 threshold read 51%. The ranking was fine, the threshold was not. This walks through checking AUC in a spreadsheet with sorted scores and the trapezoidal rule.