We want the people on-site to focus only on the work on-site.
Raising the accuracy of the "TOTTARROW" recognition AI, together
Pixie Dust Technologies is a deep-tech venture applying wave-control technology across construction DX, healthcare and more. ANOSUPO built the large volume of high-quality training data needed to push the recognition AI of its rebar-inspection DX service "TOTTARROW" one step further in accuracy.
Going one step further required large-scale, high-quality training data
Raising "TOTTARROW"'s recognition AI to a level trusted on real sites called for extensive annotation of rebar images. Handling that in-house alongside algorithm development would have drained the engineers' time — the very challenge the product itself set out to solve: letting on-site people focus on on-site work.
Pixie Dust Technologies, Inc.
They came to us and said, "We started with two categories, red and blue, but a new pattern that should be classified as green has appeared." That kind of proactivity was invaluable.
A 6-person team, color-coding rebar by type and aligning on ambiguous specs
ANOSUPO formed a dedicated six-person team, color-coding each type of rebar for identification. Where specs were ambiguous, the team checked in as needed and built the data to a consistent standard the AI could learn from, proactively surfacing new patterns from an on-site perspective — a partner working alongside them to raise accuracy, not simple outsourcing.
Completed in 2 months — vs. an estimated 4 in-house — saving ~2 months of engineer time
ANOSUPO completed the work in two months. Done in-house alongside algorithm development, it was estimated to take around four. The result: roughly two months of engineer time saved, and the development team freed to focus on its core algorithm work — supporting, from the data side, a product built to keep on-site people on the site.
Ready to achieve the same results?
Even if your requirements aren't finalized yet, that's fine. Feel free to reach out.
