Data sourcing for AI labs and data teams

Real-world data for AI,
from clinic notes
the web can't reach.

Datadhara is a data-sourcing company. We collect records that never get published online, from clinics, labs, shops and call centres, take out personal information, and license them to AI teams with written consent from every source.

Or email hello@datadhara.in. You'll hear back from me, Anuttam.

Example · Clinic notesReady to license
Domain
Outpatient clinical notes
Format
Text + scanned forms
Languages
Multilingual, incl. code-mixed
Volume
Scoped to your request
Provenance
Traceable to source agreement
{
  "note": "Pt reports  since ;
           prescribed amoxicillin 500mg",
  "patient": "[REDACTED]",
  "facility": "outpatient_clinic"
}
ConsentDe-identifiedAI-training license
Example · Shop invoicesReady to license
Domain
Wholesale & retail invoices
Format
Scans + phone photos
Labels
Line items, totals, dates
Volume
Scoped to your request
Provenance
Traceable to source agreement
{
  "vendor": "",
  "items": [{ "desc": "Rice 25kg", "qty": 4,
             "unit_price": 1150 }],
  "total": 4600,
  "tax_id": "[REDACTED]"
}
ConsentDe-identifiedAI-training license
Example · Call centre audioReady to license
Domain
Customer support calls
Format
Audio + checked transcripts
Languages
Low-resource & code-mixed
Labels
Speaker turns, intent
Provenance
Traceable to source agreement
{
  "speaker": "customer",
  "t": "00:14.2",
  "text": "order  still not
           delivered, can you check?",
  "intent": "delivery_delay",
  "phone": "[REDACTED]"
}
ConsentDe-identifiedAI-training license
Example · Lab reportsReady to license
Domain
Diagnostic lab results
Format
PDF + structured values
Labels
Test, value, unit, range
Volume
Scoped to your request
Provenance
Traceable to source agreement
{
  "test": "HbA1c",
  "value": 6.1, "unit": "%",
  "ref_range": "4.0–5.6",
  "flag": "high",
  "patient_id": ""
}
ConsentDe-identifiedAI-training license
Example · Store ledgersReady to license
Domain
Inventory & daily sales
Format
Tables / time series
Coverage
Multiple stores, daily
Volume
Scoped to your request
Provenance
Traceable to source agreement
{
  "store": "store_014",
  "date": "2026-10-03",
  "sku": "SKU-88213",
  "units_sold": 37,
  "stock_end": 112
}
ConsentDe-identifiedAI-training license
We collect from

What a delivery looks like

One record, before and after. The source keeps the original; you get a clean, labelled copy with personal details removed and a reference back to the consent agreement.

What data can Datadhara source?

We don't sell a fixed catalogue. We collect to your spec. These are the kinds of records we work with today.

CategoryExamplesFormats
HealthcareOutpatient notes, prescriptions, lab reports, discharge summariesText, scanned forms, PDFs, structured values
Business documentsInvoices, bills, purchase orders, catalogues, price listsScans and phone photos, printed and handwritten
SpeechSupport and sales calls, voice notesAudio with checked transcripts, including low-resource and code-mixed languages
OperationsSales, inventory, logistics and service recordsTables and time series
ImagesShelves, products, documents, workplacesPhotos with bounding boxes or captions

Looking for something not on this list? Ask. If we can't get it, we'll say so.

How does a data project work?

Every project starts with a small pilot, so you can check the data before you commit to anything bigger.

  1. You send a spec

    Domain, languages, format, rough volume, and what your current data is missing. We're happy to sign an NDA first.

  2. We deliver a pilot

    A small sample, collected and cleaned to your spec. You check the quality against your own evals.

  3. We scale and license

    If the pilot works, we agree volume, timeline and license terms in writing, then deliver in batches.

How are consent and personal data handled?

Data for AI training is only useful if you can defend where it came from. So that's where we start.

Written consent from every source
Each organisation signs an agreement that explicitly covers use for AI training. We don't scrape, and we don't resell data of unknown origin.
Personal information removed
Names, phone numbers, addresses, ID numbers and patient identifiers are removed before data leaves the source. Dates and locations that could identify someone are coarsened or masked.
Traceable records
Every record links back to its source agreement and collection date, so your legal team can check the chain.
Built around data-protection law
We design collection around GDPR and similar laws, and we'll work through your own compliance checklist with you. Read the full methodology or our training data rights checklist.

Questions

Something else? hello@datadhara.in

What does Datadhara do?

Datadhara sources real-world data that isn't on the public web, such as clinic notes, lab reports, invoices, call recordings and store records. We clean it, remove personal information, and license it to AI teams for training and evaluation, with written consent from each source.

Who is it for?

AI labs doing pretraining or post-training, applied AI teams fine-tuning models for a specific industry, data vendors who need licensed supply, and research groups building evaluation sets.

Where does the data come from?

Directly from the organisations that create it: clinics, hospitals, labs, retailers, wholesalers, call centres and other businesses. We collect under a written agreement with each one.

How do you handle personal data?

We remove names, contact details, ID numbers and patient identifiers before data leaves the source, and mask dates or locations that could identify someone. Each dataset card records how this was done.

Do you have off-the-shelf datasets?

Not yet. We collect to your spec. Tell us the domain, languages, format and volume, and we'll scope a pilot. Exclusive and non-exclusive licenses are both possible.

Is the data licensed for commercial model training?

Yes, that's the default. Licenses cover training and evaluating commercial models. Exclusivity, duration and field of use are agreed per project.

Which languages do you cover?

It depends on the sources for each project. Alongside English, we work with low-resource and code-mixed languages in both text and speech. Tell us which languages you need and we'll confirm before scoping.

What is a dataset card?

A short document that ships with every delivery. It describes what the dataset contains, where it came from, how it was collected and labelled, how personal information was removed, and its known gaps.

How long does a pilot take, and what does it cost?

Both depend on the domain, format, volume and how much labelling you need. We agree the timeline and price when we scope the pilot, and keep it small so you can judge quality before spending more.

Tell us what your models are missing

A few lines is enough: the domain, the format, and roughly how much. We reply within one working day.

Datadhara is new. Right now it's me talking to clinics, labs and businesses about their records, and to AI teams about the data they can't find.

If you're building something that needs data like this, I'd like to hear about it.

Anuttam Anand founder, hello@datadhara.in