Teach Nanonets the exact fields Registry needs from one document type.
Training a model for one document type teaches Nanonets the exact fields Registry needs and improves extraction accuracy. Starting with passports keeps the pilot narrow and consumes fewer credits.
Turn the pilot into an evidence-based decision on whether to scale.
Get the extracted data into the systems Registry actually works in.
Prove the accuracy on one document type before scaling to others.
Choose the single document type to pilot. Passports are the highest-volume and most consistent.
Define the exact fields Registry needs from the chosen document type.
Upload a set of samples covering the variations you see.
Label the fields on the samples so the model learns them.
Train the model. Training consumes credits.
Test the trained model on documents it has not seen.
Refine with more samples if accuracy is short.
**Scenario**: Registry needs to extract [field list] from passports. **Step 1**: Choose passports as the pilot document type. **Step 2**: Define required fields: - [Field A] - [Field B] - [Field C] **Step 3**: Upload [number] sample passports covering [variation A], [variation B], and [variation C]. **Step 4**: Label each field on all samples. Draw a box around [field name] on each document and assign the label. Repeat for the other fields. **Step 5**: Train the model in Nanonets. **Step 6**: Upload [number] new passports not in the training set. Compare extracted values to the actual text. **Step 7**: If accuracy is short, add more samples that cover the errors you observed, label them, and retrain.
Define the fields Registry needs from the chosen document type, upload and label a representative sample set covering the variations seen, train the model, then test on unseen documents.
Pull the key clauses and signatories out of a long document automatically.
Cross-check extracted values against what Registry already holds.