wonderlab

the lab where any question becomes a real research bridge.

ask anything you wonder about. iris answers in plain words, draws what she's saying live on a 3D whiteboard, and ties it to a real open scientific challenge — so a "why does cheese melt?" reads as both a kid answer and a frontier protein-stability problem.

why does cheese melt? cold — pieces hold tight heat ↑ warm — pieces slip past
researchers say protein-fat
thermal disordering
the product

a translation layer between
everyday curiosity and real research

wonderlab speaks two languages at once. visible answers and pictures stay in plain words. the technical version lives in a research layer underneath — same idea, real terminology, linked to the actual benchmark.

what visitors ask — in plain words

  • why does cheese melt?
  • why is the sky blue?
  • why can't a cooked egg go back to runny?
  • how does a magnet hold the sun?
  • why does medicine know which spot to fix?

what researchers see — same answer, tied out

  • protein-fat thermal disordering · Hugging Science · stability
  • Rayleigh scattering, λ⁻⁴
  • irreversible cross-linking via disulfide bonds
  • magnetic confinement geometry · fusion / stellarator
  • drug–target interaction · Therapeutic Data Commons
how it works

three jobs, one researcher

📖

translate

iris reads who's asking and replies at their level. kid words on every visible surface. the technical version sits behind a "researcher words" hover.

✏️

draw

every answer comes with a hand-drawn SVG illustration on the whiteboard — before-vs-after, leader lines, plain-words labels. the picture and the words are one teaching unit.

🔗

link

iris ties each answer to a real open challenge — Hugging Science protein stability, TDC drug discovery, Materials Project, AlphaFold DB. one click takes you to the upstream dataset.

bring your own model

wonderlab is 100% static — no server, no logging, no proxy. your question goes from your browser straight to whichever model you pick. your API key (when you need one) lives only in your browser.

Gemma 4 · open weights · free Claude (Anthropic) Gemini (Google) WebLLM (your GPU) LM Studio (localhost) 23 languages kid · curious · grad depths
built for

five-year-olds and grad students
read the same room

the depth knob picks the voice. the translation rule never breaks.

🧒

kids

plain words, hand-drawn pictures, one aha per answer.

🤔

curious adults

real explanations, clear metaphors, no condescending.

🎓

classrooms

save any answer as a printable notebook PDF. embed in a blog.

🔬

researchers

open the "researcher words" reveal. follow the benchmark link.

why this exists

the missing translation layer

Hugging Science put open scientific datasets, models, and benchmark leaderboards on Hugging Face — fusion stellarators, antibody developability, drug discovery, genomics. Their pitch literally complained: "You shouldn't need to scrape arxiv, run your own wetlab, fight a custom HDF5 parser, build a fusion stellarator, and beg for compute before you've trained a single epoch."

wonderlab fills one layer their roadmap doesn't: the friend who turns a normal question into a real scientific challenge submission, and turns the result back into language the asker understands. the 5-year-old framing is the stress test — real users are anyone non-expert with intuition.

— a small bridge from "i wonder…" to "this is a real open problem."

open the lab. ask anything.

iris is waiting at the whiteboard. she answers in plain words and draws what she's saying. one click takes you to the real benchmark.

open the lab →