Brazilian Personal Data Redaction for Portuguese Text (LGPD)
This model package finds Brazilian personal data in Portuguese text and can replace, mask, remove, or only report each finding. It runs as a SageMaker real-time endpoint or batch transform job in your account, with network isolation. The container makes no outside calls.
The product name includes LGPD because the listing is aimed at Brazilian personal data. Running it does not certify LGPD compliance, and a successful call does not mean every personal-data item was removed.
- Delivery
- SageMaker model package in your own AWS account
Quick start
Full guide →What it finds¶
The listing names these types: people's names, CPF, CNPJ (including the alphanumeric form, with check digits), RG, CNH, PIS/PASEP, voter ID, OAB, RENAVAM, vehicle plates (old and Mercosul), bank branch and account, card numbers, PIX keys, phones, emails, CEP, addresses, and dates of birth.
Structured identifiers are found by rules written for Brazilian formats. Names come from the GLiNER2 privacy filter model (Apache-2.0, used as published), with filters for role words, titles, and place names. Addresses and dates of birth are found by rules. The listing states that no customer data was used to build it.
What it returns¶
The default action replaces each finding with a placeholder derived from a secret key you send with the data. The same key and the same value always produce the same placeholder, so cleaned records can still be joined. You can instead mask, remove, or only report, including a different action per type. The response lists each finding with its type and its position in the original text. Original values are returned only when you ask for them.
Replacing a value with a placeholder is pseudonymisation, not anonymisation.
Keep the key secret. Without a key, the model creates a temporary key at
start-up and every response says "key": "temporary".
Recommended instances¶
The listing recommends ml.m5.large for a real-time endpoint and
ml.c7i.xlarge for batch. It also says a new account may need a quota
increase for ml.c7i batch transform, and that ml.m5.xlarge and
ml.c5.2xlarge work for batch without one in most accounts. The
sample notebook
runs its sample endpoint and its three sample batch jobs on ml.m5.large.
Billing¶
Software is billed per instance-hour, for a running endpoint and for a batch job. The public offer lists these software rates. Confirm the offer before you subscribe.
| Instance | Real-time | Batch |
|---|---|---|
ml.m5.large |
USD 0.22 | USD 0.22 |
ml.m5.xlarge |
USD 0.44 | USD 0.44 |
ml.m5.2xlarge |
not offered | USD 0.81 |
ml.c5.xlarge |
USD 0.47 | USD 0.47 |
ml.c5.2xlarge |
not offered | USD 0.88 |
ml.c6i.xlarge |
USD 0.55 | USD 0.55 |
ml.c7i.xlarge |
USD 0.62 | USD 0.62 |
ml.c7i.2xlarge |
not offered | USD 1.01 |
The offer includes a free trial. Confirm its length and what it covers. The
sample notebook says the software price is not charged during the free
trial. SageMaker infrastructure charges are separate either way. The public
notebook estimates about 15 minutes of ml.m5.large and under USD 0.10 of
AWS infrastructure for its sample run. The repository README estimates about
20 minutes and under USD 0.10 of AWS charges for the full notebook.
Walkthrough¶
Watch on YouTube. The listing says the interface in the video is an illustrative integration, not a product UI, and that the examples are fictional while the model responses are from the published model.
Where to go next¶
- Getting started: subscribe, create the model, run an endpoint or a batch job, and delete the test resources.
- Usage: JSON, JSON Lines, CSV, plain text, the key, and options.
- Limits and evaluation: what the published measurement covers.
- Support: how to write in, and what not to send.
- Sample notebook and data.
- Shared deployment steps.
Last updated November 11, 2026 · Suggest an improvement