Zapier Nanonets Data Extraction Setup
Budget: $250 – $750 CAD
I need a Nanonets OCR model trained so it can live inside a Zapier workflow and automatically feed Zapier Tables. The source material is a steady stream of Word documents along with occasional handwritten images. From the Word files the model must reliably capture names and full postal addresses; from the handwritten images it should grab email addresses that people jot down.
Here is the flow I have in mind: once a new file lands in the designated cloud folder, Zapier calls the Nanonets endpoint, gets back the parsed data, and writes each record—name, street, city, state, ZIP, and email—sample descriptions and number of samples into its own row inside an existing Zapier Table.
You’ll receive anonymized samples for both document types, plus access to my Nanonets dashboard and the Zapier account. The end-result I expect is:
• A trained and published Nanonets model with clearly labeled fields
• A Zapier Zap that triggers on new files, calls the model, and maps the returned text into the correct table columns
• A short hand-off doc or loom video showing how to add new samples and retrain if accuracy dips
Everything has to run hands-free once delivered; if the model scores below 90 % on the test set, we’ll iterate until it clears that bar.
Here is the flow I have in mind: once a new file lands in the designated cloud folder, Zapier calls the Nanonets endpoint, gets back the parsed data, and writes each record—name, street, city, state, ZIP, and email—sample descriptions and number of samples into its own row inside an existing Zapier Table.
You’ll receive anonymized samples for both document types, plus access to my Nanonets dashboard and the Zapier account. The end-result I expect is:
• A trained and published Nanonets model with clearly labeled fields
• A Zapier Zap that triggers on new files, calls the model, and maps the returned text into the correct table columns
• A short hand-off doc or loom video showing how to add new samples and retrain if accuracy dips
Everything has to run hands-free once delivered; if the model scores below 90 % on the test set, we’ll iterate until it clears that bar.