Build PII detection training sets — span by span.Read the label spec
34 labels, mapped to ner_typeAPSIS spec

[ RGPD-ALIGNED SPAN ANNOTATION ]

Turn raw text into PII training data.

Paste a document, let the deterministic layer pre-mark every email, IBAN, NIR and phone number, then correct the rest by keyboard. Export a training-ready JSONL in one click.

/* From blank dataset to exported JSONL */

$ bun dev → /datasets → INSERT DATA → EXPORT

[ WHAT THE DETERMINISTIC LAYER COVERS ]

$LABELS

34

$AUTO_DETECTORS

14

$CHECK_DIGITS

5

[ HOW IT WORKS ]

01PRE-ANNOTATE

Regex plus real check digits — IBAN mod-97, Luhn, the NIR key — mark the structured identifiers before you touch the keyboard.

typevalue
EMAILF0DD78E1••••••••
IBAN9C41A0B2••••••••
NIR2E77BD10••••••••
PHONEA5310FCC••••••••
02SELECT AND LABEL

Drag over any span and press one key. ner_type, token family, RGPD regime and the masking default are derived from the label.

03EXPORT

One JSONL line per record, deduplicated by (entity_type, original) and ordered by first appearance, exactly as the training prompt promises.