How we build

How we build GiftyWow

Decisions, mistakes, and lessons from building a personalised gift recommendation engine. For product people, AI builders, and experience designers who care about subjective quality.

Articles

How we build 11 min read

Standard AI Evals Assume One Right Answer

LLM-as-a-judge for personalised gifting: why one-expert benchmarks fail, and how we built in-product labs, a proprietary rubric, and scalable evals with no single correct answer.

Octavia Maddox, Founder · June 16, 2026

See what we are building

Upload two photos and swipe through gift ideas matched to both of you. Free to try.

Start Matching

See what we are building

Upload two photos and swipe through gift ideas matched to both of you. Free to try.

No app needed. Opens right in your browser.

GiftyWow app: From and To photo upload with privacy note and Why Photos preview

Upload photos
to skip long forms.

Review and edit
the vibes.

GiftyWow app: swipe gift card with Save sticker, personalised tumbling tower product, tags, and actions

Swipe your
personalized gift edit.

Articles

Standard AI Evals Assume One Right Answer

See what we are building

See what we are building