New · batch upload

Handwritten lab pages, digitized and checked against themselves.

Drop a photo of a prep sheet, or a folder of them. Each reagent table transcribes itself, then the numbers get verified three independent ways. You review the cells that disagree instead of all forty. Finished pages stack up in a sidebar you can sort so the flagged ones come first.

Try the demo Run it in your lab

What a flagged page looks like

Reagent	MW g/mol	Equiv	mmol	Mass g
salicylic acid	138.12	1	10	1.381
acetic anhydride	102.09	1.5	15	1.531
sodium acetate	82.03	0.5	5	0.510

Checks

2 pass1 flagged

sodium acetate

written 0.510 g · computed 0.410 g

second read: 18/18 fields confirmed ✓

The transcription is faithful, so the mismatch is on the paper.

Three checks, three blind spots covered

Arithmetic

Mass in grams should equal mmol times molecular weight. Every row is recomputed and compared against what is written. Catches transcription errors in numbers.

Database

Written molecular weights are checked against PubChem, with hydrates handled properly. A 6H2O salt is looked up anhydrous and corrected by six waters. Catches wrong values used on the page itself.

Second read

A separate pass re-reads the image without trusting the first transcription and reports only the fields where the two readings disagree. Catches misread names and missed rows.

Nothing is corrected automatically. The extraction transcribes exactly what is written, even when it is wrong, which is what makes the checks worth anything. Flagged cells point back at the photo and a human decides.

Two ways to run it

Demo

Try it in a browser, no setup.

Up to 3 pages at a time, or a folderNew
All three verification passes
JSON and CSV export
History kept in your browser only
A few runs per hour, or bring your own key for noneNew

Extraction runs through Anthropic's API, so keep sensitive pages off it.

Open the demo

Self-hosted

In dev

Your machine, your key, your data.

Batch size is yours to setNew
No shared key and no shared rate limit
Local vision model option, nothing leaves your network
All three verification passes
JSON and CSV export
MIT licensed, fork and modify freely

Node and a terminal for now. A one click deploy and a Dockerfile are still to come.

Setup instructions

Running this for a lab and want help setting it up? Get in touch.

Run it yourself

Clone it, add a key, and it runs on your machine. Point EXTRACTION_BASE_URL at a local vision model instead and no page ever leaves your network.

git clone https://github.com/tornike14/labshot
cd labshot
npm install
cp .env.example .env.local   # add your ANTHROPIC_API_KEY
npm run dev