How it works

Raw sessions aren't training data. They become it in three steps.

Each step is where most of the value, and most of the risk, sits.

A 19th-century engraving of the Rock of Gibraltar from the sea
The steps
01

Select

Most sessions are routine. We look for the ones where a model failed or an expert corrected it on something hard, and drop the ones that only failed because something else broke.

02

Rights & redaction

Rights reviewed with counsel before anything moves. Redacted with two independent passes and a human spot-check, and the source approves every sample before anyone sees it.

03

Verify

In code, a test tells you if the work is right. In a contract or a financial model, it usually can't. We rebuild the task with its corrections and write the grader. We try to break every grader with shortcut answers before it ships.

What a task looks like

One contract revision, one expert correction, one grader that checks it.

Instruction

Revise the limitation-of-liability clause: cap liability at 12 months of fees paid, and exclude breaches of confidentiality from the cap.

Model draft

Caps liability at fees paid in the prior 12 months. No carve-out for confidentiality.

Expert correction

“No, not like that. Confidentiality breaches sit outside the cap, and nothing else in the agreement changes.”

Grader
  • ✓ Cap equals 12 months of fees
  • ✓ Confidentiality breaches excluded from the cap
  • ✓ No other clause modified

Illustrative example. Not real client data.

Contact

We’re early, and talking with AI companies and lab data teams now.

hello@pillarsdata.com

Started by Patrick Ojo · X · LinkedIn · Dallas, Texas