A dedicated human data team for your models
Deepen AI's trained raters evaluate your models, red-team them, rate responses for post-training and record agent demonstrations. The same people every round, with a weekly report.
Rate quoted in your plan A plan within 2 working days No sales call
The human data behind your models
From the first eval to post-training data, one team that learns your rubrics and keeps them.
Why a dedicated team
Human data is only as consistent as the people producing it. A team that stays on your project keeps your rubric the same from round to round.
| Topic | An open pool of raters | A dedicated Deepen team |
|---|---|---|
| Who rates | Whoever picks up the task | The same trained raters, round after round |
| Rubric changes | Re-explained to whoever is next | Trained into the team once, with the date noted |
| Unclear items | Each rater decides alone | The team lead decides, or asks you |
| Consistency | Varies as the pool changes | Agreement checked and reported every week |
Where the team works
In the platform your researchers already use, on accounts you control.
In your annotation or eval platform
Raters work in your own tools, so items, rubrics and results stay where your pipeline expects them.
Through a shared sheet or bucket
Your pipeline writes items to a sheet or bucket, and rated data is written back in your schema.
How work reaches the teamYour data and model outputs stay confidential. We work under NDA, with the access you set. Deepen AI holds SOC 2 Type II, ISO 27001, ISO 9001 and TISAX.
How rating quality is kept
Raters calibrate on known items before live work, a share of items is rated twice to check agreement, and QA re-checks a sample. You see agreement and accuracy every week.
Every reviewer's accuracy is measured
Items you've already answered are hidden in the real work, so we know how accurate each person is.
Nobody guesses on unclear cases
The team lead decides, or asks you. Your answer joins the guidelines.
You see the results every week
A weekly report shows work done, QA results and what's improving.
How we start on your work
No sales call. Start with one project: an eval round, a red-team sprint or a batch of preference data.
What to bring
- Your rubric or rating guidelines
- A set of items you've already rated
- Access to your platform, or the format your pipeline reads
- Your volumes and your timeline
- 5 minutesTell us the task
What you need rated or recorded, how much, and by when.
- Within 2 working daysGet a plan and a price
Raters, hours, the rate, start date and the total price, by email.
- 2 weeksRun the pilot
We train raters on your guidelines, check they agree, rate real items and send two weekly reports. Then you decide.
Pricing
Evaluation, preference data and agent data are quoted in your plan, because the rate depends on the expertise and the task. Your plan shows the rate, the team size, the hours and the total price before you pay anything.
Set before you pay. Your plan shows the rate, team size, hours and total price.
Starts with a two-week pilot on your real items, paid upfront, with two weekly reports.
Then you decide. After the pilot, you choose whether to continue.
Questions
Do the same raters stay on our project?
Yes. Your team stays on your work, so your rubrics and edge cases aren't relearned every round.
Can raters with domain expertise work on our data?
Tell us the domain when you request a plan. The plan says who will rate your data, and the rate, before any work starts.
Can the team grow for a large training run?
Yes. Tell your team lead ahead of time. New raters are trained on the same guidelines and calibrated on known items before they rate live work.
Which formats do you deliver?
Your schema, such as JSON lines or CSV, with the fields you set, including the reason behind each rating if you want it.
How is our data used?
Only for your project. Your prompts, outputs and rated data stay under NDA, with access you control.
Can we start small?
Yes. The two-week pilot is a good size for one eval round or one batch of preference data, with two weekly reports before you decide.