Before you trust the forecast,
trust the numbers behind it.
A financial-data problem, solved at the input. I built and published numbr to interpret numbers written in text and convert them into usable Python values. The live scenario shows how that capability fits into invoice review.

Model outputs are only as credible as their inputs. In the scenario below, numbers from fictional supplier invoices are extracted, validated and organized for financial review—without pretending that unchecked values are ready for payment.
See the problem, then interact with the working example.
Use numbr to interpret text-based numerical values; run additional workflow checks against fictional invoice fields.
Structured, reviewable inputs so a finance team can investigate exceptions before building a payment plan.
Demonstration scenarios and financial records are illustrative. The published Python library and the surrounding demo workflow are identified separately.
Engine off · connect to run
Connected
Turn invoice text into numbers finance can review.
No retyping. Choose a supplier invoice and run its source text through numbr. See the numbers it extracts, where they came from, and which invoices are safe to include in a payment plan.
PYTHON ENGINE / LIVE ACTIVITY Waiting for Start engine
The component
I built numbr, a Python library that finds numbers in text and converts written forms such as “twenty-four” into 24.
The finance context
Finance teams need reliable numbers without retyping supplier documents. Extracted quantities, rates and payment terms can feed a payment plan; unclear values need a person’s review.
In this live demo
Choose a fictional invoice and click Extract this invoice. Follow each source highlight and its real Python result. My demo then checks the totals and plans the payment.
Fictional invoices · actual Python package · labeled sample text, not image scanning.
November 7, 2026
This is the supplier’s document, not an input form. Run extraction above. The reading guide reveals each Python result and highlights its source here, one field at a time. Pause or select Next field to read at your own pace.
Source invoice figures are shown as billed. Extracted quantities, rates and terms are validated separately before the payment plan is calculated.
Inspect the labeled fields sent to Python
Loading invoice fields…
See the actual Python calls +
Run extraction to inspect the actual Python calls.
Explore all number conversions and text extractionOpen the full playground +
Try the general-purpose numbr toolkit.
The component
The same numbr package converts numbers between digits, words, ordinals and Roman numerals, and extracts numbers from short text.
The finance context
Enter a number or sentence below. The Python API runs the actual package; unsupported conversions are explained before you submit.
Roman numerals support whole numbers from 1 to 3999. Examples outside that range are disabled.
import numbr
numbr.Cast("forty-two", "Cardinal Number")Roman output supports integers 1–3999. Unsupported inputs and conversions return the package's feedback.