Skip to main content
Shipped 2026-03-21 · AI Most OCR is built for receipts and street signs. That’s not what you’re photographing at the bench. We’ve upgraded the photo-extraction pipeline to read lab content well — handwritten notebook pages, instrument readouts, gel labels, reagent vials.

​
What changed

  • Model upgrade to Gemini 2.5 Pro (and 3.1 Pro Preview where available) — substantially better on scientific vocabulary, units, subscripts, and chemical formulae.
  • Lab-specific prompt — the extraction prompt knows about reagents, protocol parameters, units (mM, µmol/L, °C), and common lab notebook conventions, so it makes fewer guesses on ambiguous characters.
  • Confidence reporting — when Bower can’t read part of an image clearly, Bird flags the specific region and asks you to verify before the values are committed to your note.

​
Why this matters

OCR that misreads “240 µmol/L” as “240 µmol/I” silently corrupts your data. The new pipeline is honest about its uncertainty — surfacing low-confidence regions for human review rather than guessing.