ABOUT PIKFIND

Built at a real festival, by an AI-native team.

PikFind lets event participants find their own photos among thousands with a selfie or a bib number — no app, no sign-up. It is built and operated by kinoflare, a software company in Seoul.

WHY IT EXISTS

We ran SIDF, the Seoul International Dance Festival. 5,186 photos, three photographers, one weekend — and handing them out took months. PikFind was built to change that order: organizers upload once, participants find their own photos the same day.

IN THE FIELD

photos processed with this search technology: 1,000,000+
photos delivered at SIDF 2026: 5,186
photos at an event hosted with the Embassy of Colombia: 1,128
languages supported: 5

TECHNOLOGY

Face-embedding search

InsightFace (buffalo_l) 512-dimensional face embeddings match a selfie against every face in the event. Selfies and their features are deleted as soon as the search ends.

Bib-number OCR

RapidOCR reads race bibs on onnxruntime, so marathon runners can search by number and fall back to their face when the bib is hidden.

Scene clustering & auto-labels

CLIP image embeddings group photos by place and moment; a vision-language model suggests names for each group.

GPU burst processing

Large uploads burst onto cloud GPUs and spill over to CPU workers, so a 30,000-photo event is searchable quickly.

HOW WE BUILD

kinoflare is model-agnostic. We develop PikFind with several AI models side by side — Claude (Claude Code) by Anthropic, Codex by OpenAI, Gemini by Google and Grok by xAI — choosing the right one for design review, coding, testing, research and multilingual copy, with humans making the final call.

COMPANY

PikFind is a product of kinoflare (sole proprietorship, Republic of Korea · Business Registration No. 542-47-01056 · founded 2023 · CEO Geunhyeong Kim). kinoflare also develops NULX, video infrastructure with AI captioning.

About kinoflare ↗ · Contact: support@pikfind.com