It’s locked up
Support logs, records and documents hold the signal models need. Personal details keep them out of training.
Data for AI · Since 2018
Pluto turns raw, sensitive and messy data into datasets AI teams can use: extracted, standardized and de-identified. We also license data we’ve collected ourselves.
Backed by investors who have held leadership roles at
8 years
of computer vision and data research
96%
F1 score on our first fully benchmarked system
Your cloud
our pipeline runs in your environment, so your data never leaves it
The problem
Support logs, records and documents hold the signal models need. Personal details keep them out of training.
Blacking out names and numbers leaves holes. A model can’t learn much from text that’s half missing.
Every new source, format and edge case becomes someone’s maintenance job. Volume stalls.
What we do
Your sources
Pluto pipeline
01
Extract
Usable fields pulled from raw sources.
02
Standardize
One schema, deduplicated and labeled.
03
De-identify
Personal details swapped for stand-ins.
04
Review
Checked by our team before release.
You receive
See it work
Raw input · scanned invoice
NORTHWIND SUPPLY CO Invoice no. A-20931 Date 04/11/26 Bill to: Harbour Café, 12 Pier St 2 x espresso grinder ......... 1,040.00 1 x service plan ............... 208.50 TOTAL DUE CAD 1,248.50 net 30
Structured output
{ "vendor": "Northwind Supply Co", "invoice_id": "A-20931", "date": "2026-11-04", "line_items": 2, "total": 1248.50, "currency": "CAD", "terms_days": 30 }
Same person, three systems
| Source | Name | Date | Phone |
|---|---|---|---|
| CRM | SMITH, J. | 2026-03-04 | (204) 555-0110 |
| Billing | Jane Smith | 4 Mar 26 | 204.555.0110 |
| Support | jane smith | 03/04/2026 | 2045550110 |
One clean record
{ "name": "Jane Smith", "first_contact": "2026-03-04", "phone": "+1 204 555 0110", "sources": ["crm", "billing", "support"] }
3 records merged into 1. Formats unified.
Support chat · entities found
Agent: Thanks, Tom. I see order #88214 shipped to 41 Maple Ave, Brandon. Is tom.reyes@mail.com still the best email?
Replaced
| Found | Type | Becomes |
|---|---|---|
| Tom | NAME | Ravi |
| #88214 | ORDER_ID | #30577 |
| 41 Maple Ave, Brandon | ADDRESS | 9 Elm Rd, Selkirk |
| tom.reyes@mail.com | ravi.k@mail.com |
Illustrative examples. All names and details are fictional.
Why Pluto
Personal details are replaced with realistic stand-ins, so text keeps its shape and meaning.
Documents, logs, feeds and images. Inconsistent formats are the normal case, not the exception.
Output matches the spec you give us, ready to drop into your pipeline.
We’ve never outsourced labeling. The people who built the pipeline check the output.
Eight years of computer vision research, applied to extracting and labeling data at scale.
Pricing scales with how much data we process for you.
Our data
First-party data collected through our own platform, available to license.
Parallel
Parallel is our consumer fashion platform, used by 2.5 million people a year. It’s where our technology was built and tested, and the source of the data we license.
Visit joinparallel.io →Buy-button click rate on product pages
Powered by 12 live size recommendation models.
Who we work with
Real-world datasets prepared for pre-training, fine-tuning and evaluation.
Processing for the datasets you broker, so they’re ready to sell.
De-identification built into the pipeline before data reaches a model.
A privacy layer for clients running private inference.
Company
Pluto Ventures started as a computer vision company. Eight years of research taught us how to collect, label and clean data at scale. Now we do that work for others.
FAQ
Instead of blacking out personal details, we replace them with realistic fictional ones. A name becomes a different name, an address a different address. The text stays natural, so it’s still useful for training.
Yes. That’s what it’s for. Datasets leave our pipeline ready for pre-training, fine-tuning or evaluation.
By volume. Pricing scales with how much data we process for you.
Our first fully benchmarked system scores 96% F1, a measure that balances how much sensitive data is caught against how often clean data is wrongly flagged. We’re benchmarking each remaining system the same way and will publish results as they’re complete.
Send us a description of your source data and what you need it to become. We’ll scope it and come back with timing and price.
Tell us what you’re working with.