vbits

Synthetic data

vbits

Realistic, privacy-safe synthetic data, built to order.

View datasetsRequest a set

01/Catalog

Datasets

Text for language models, transactions for fraud models, and healthcare records. Listed sizes are the launch specification. Every set is synthetic.

Sold on Datarade. This site does not take payment. A buy button stays off until that dataset has a listing URL.

01

AI training text

Customer-support chat dialogues and instruction sets for language models.

Customer support dialogues

1,200,000 turns

Multi-turn customer and agent chats for training and evaluating support models.

conversation_id · turn_index · role · text · intent · resolved · language

JSONL · CSV · Parquet

Schema

Placeholder — listing URL not set

Instruction and evaluation sets

300,000 examples

Instruction, input, and output examples with rubrics, split for training and evaluation.

example_id · split · instruction · input · output · rubric · difficulty

JSONL · CSV · Parquet

Schema

Placeholder — listing URL not set

02

Bank and card transactions

Transactions with fraud labels, for training and testing fraud models.

Fraud-labeled transactions

5,000,000 transactions

Synthetic card and bank transactions with fraud labels, for fraud-model training and testing.

transaction_id · account_id · event_time · amount · currency · mcc · channel · is_fraud · fraud_type · split

CSV · Parquet · JSONL

Schema

Placeholder — listing URL not set

03

Synthetic healthcare

Patient records, claims, and patient flow. Clinical sets are ICD-10 and CPT coded. No real patients.

Patient records

750,000 records

Synthetic patient records with ICD-10 diagnoses. No real patients.

patient_id · birth_year · sex · encounter_date · icd10_code · diagnosis · care_setting

CSV · Parquet · JSONL

Schema

Placeholder — listing URL not set

Insurance claims

2,000,000 claims

Synthetic claims coded with ICD-10 and CPT. No real patients.

claim_id · patient_id · service_date · icd10_code · cpt_code · billed_amount · paid_amount · status

CSV · Parquet · JSONL

Schema

Placeholder — listing URL not set

Patient flow

500,000 encounters

Synthetic encounter pathways: arrival, department, length of stay, and disposition.

encounter_id · patient_id · arrival_time · department · acuity · length_of_stay_min · disposition

CSV · Parquet · JSONL

Schema

Placeholder — listing URL not set

02/Process

How it
works

From a specification to a labeled dataset.

  1. 01

    Request

    Share the subject, the fields, the volume, the format, and the use case.

  2. 02

    Generate

    A new dataset is created to that specification. It is fully synthetic.

  3. 03

    Validate

    The schema and labels are checked. A generation note records the methods, the algorithms, and the limitations.

  4. 04

    Deliver

    Files ship marked as synthetic, with that note. The sale is completed on Datarade.

Synthetic, on the record

  • Every dataset is 100% synthetic. It is not copied from real customers, cardholders, or patients.
  • Each product is labeled as synthetic in the title, the description, and the file metadata.
  • Delivery includes a generation note: the methods and algorithms used, the schema, and known limitations.

Datarade's synthetic data policy asks for the same three things: disclose synthetic data in the title and description, label it clearly, and document the generation methods and algorithms. Publishing policies, section 4.4.

03/Custom

Request

Describe the set you need. There is no checkout on this site. When a dataset is ready, it is sold on Datarade.

Do not include real personal data in this request.

PlaceholderThis form is not connected. Set NEXT_PUBLIC_FORM_ENDPOINT to a Formspree-compatible URL and rebuild. Nothing is sent until then.