NSK AI
Products
·
Zerra

An AI that learns your people, then teaches them.

Zerra is an adaptive learning platform for upskilling. It estimates what each person can already do, routes them to the next thing worth learning, and proves the gain. The same engine runs your in-house assessments, curated with our own tools.

Four things happen between a person arriving and a skill being signed off.

Scroll to follow them
01  Ability

A person is a point, not a level.

Every skill in your organisation is an axis. Zerra starts by placing the learner on it, from the work they already do.

02  Estimate

Each answer narrows the guess.

Items are picked where the answer carries the most information. The band around the estimate tightens as the evidence arrives.

03  Route

The path re-plans as they move.

Prerequisites hold the graph together. Zerra walks the shortest route through it that still clears your standard.

04  Evidence

Mastery you can audit, per person.

One cell per learner per skill, with the items behind it. This is the same view your in-house assessments write into.

P(correct) = σ(θ − b)
The mechanism

Five steps, and none of them are a course catalogue.

01
Baseline

A short adaptive sitting, on tasks drawn from your own work.

18–24 items. No fixed pass mark; the estimate is a range with a confidence band.
02
Diagnose

Wrong answers are read for cause, not counted.

Misconceptions are tagged against the skill graph, so the same gap is not taught twice.
03
Route

The next step is chosen after every answer, not at enrolment.

Prerequisites are hard edges. Everything else is negotiable and gets skipped when already held.
04
Practise

Work in the tools they use on Monday, not a simulator.

Exercises run against your stack and your data policy, with an engineer reachable in the loop.
05
Verify

Competence is signed off on evidence, not attendance.

Every claim links to the items that support it, exportable for audit.
In-house assessment

Assess your own workforce, on your own standard.

Build as many item banks as you have functions. Each one is curated against your competency framework, calibrated on your cohorts, and reported with the agreement rate behind it. Use them for hiring screens, internal certification, promotion evidence and regulatory sign-off.

Unlimited item banks, one per function or role
Calibrated difficulty, published with each item
Reviewer agreement reported for every bank
Results exportable, per person, team and skill
Your item banks · calibrated difficulty
hover a column
easier
b ∈ [−2.4, 2.4]
harder
Adapts per person

Two people in the same team can be on completely different paths by the second week, and both finish on your standard.

Adaptive engine
Built on your material

Your runbooks, tickets and code become the content. Nothing generic is taught where something specific exists.

Content pipeline
Assessment you can defend

Calibration, agreement and item history are all visible. A result survives a conversation with legal.

Evidence layer

The rest of the system

Own your own AI future.

Train your people on the system you actually run, and test them on it.