Home / Sample data
Sample data
These are free CSV samples that show the structure, labels and de-identification of each type of data Datadhara licenses. Every row is synthetic, so you can share them with your team or run them through your pipeline before talking to us. The samples are released under CC BY 4.0.
Download the samples
Every record in these files is synthetic. They were generated to show structure, labels and de-identification. None of them comes from a real person, clinic or business. Licensed deliveries contain real records collected under written agreements.
| Sample | What's in it | Rows | File |
|---|---|---|---|
| Clinic notes | Outpatient notes with age band, sex, visit month, specialty, free-text note, ICD-10 code and label, and medications. | 16 | CSV |
| Lab reports | Structured lab results: test name, value, unit, reference range, high/low flag and method, one row per test. | 32 | CSV |
| Invoices and bills | Line-item labels for receipts and bills: item, quantity, unit price, line total, tax rate, invoice total and document type, with the vendor name redacted. | 31 | CSV |
| Call-centre transcripts | Turn-level transcripts of support calls with speaker, timestamps, language tag (including code-mixed Hindi-English), intent and sentiment. | 25 | CSV |
| Retail sales ledger | Daily store-level sales and stock: SKU, category, units sold, price, revenue, opening and closing stock, and a reorder flag. | 42 | CSV |
What to look for
- De-identification. Names, phone numbers and IDs are replaced with stable pseudonymous IDs (for example
pt_0ee42548) or[REDACTED]. Exact dates are coarsened to a month and ages to a five-year band. - Provenance. Every row has a
consent_refthat points to the source agreement and its version. In a real delivery, that reference resolves to the signed agreement. - Source format. Each record says what it was captured from (handwritten and scanned, phone photo, POS export, call audio), because models behave differently on each.
- Messiness kept on purpose. Abbreviations, code-mixed speech and crossed-out lines stay in the data. Cleaning them away would remove the reason to use real-world data.
Every delivery comes with a dataset card covering collection, labels, de-identification and known gaps. See the methodology for how that works, and how provenance is recorded.
The same files are on Kaggle, GitHub and Zenodo (DOI 10.5281/zenodo.24349905, for citation).
Questions
Are the samples real data?
No. Every row is synthetic, created to show the format. Licensed deliveries contain real records collected with written consent from the organisation that holds them.
Can I get a real sample?
Yes. Paid pilots start small, with a few hundred records from real sources, delivered with a dataset card, so you can test them before committing to volume.
Can the format match our schema?
Yes. The columns here are a default. Deliveries can follow your schema, label set and file format.
Tell us what your models are missing
A few lines is enough: the domain, the format, and roughly how much.
Request data →