Viceron STATLAB · analytical platform

Ask for the analysis. Get the code.

Describe what you want to know in a sentence. STATLAB chooses the method and says why, checks the assumptions that method depends on, runs it against a frozen version of your data, and hands back the result, the method paragraph and the R or Python that produced it. You can check every answer yourself.

R & Python outputAssumption checksSeeded & versioned Statistician escalationRuns in your perimeter Our models or NVIDIA NemotronNVIDIA OpenShell runtime
Book a walkthrough→ Watch four questions↓
● The premiseWhy every answer comes with its code

A statistic you cannot reproduce is a rumour with a confidence interval. So the platform's real output is the script behind the number.

There is a version of this product that just answers questions, and it would be easier to sell. It would also be unpublishable, because you, your co-authors and reviewers would have no way to check what it did.

STATLAB is built the other way round. Every run emits standard R or Python using standard packages, the frozen dataset version it ran against, the random seed, the package versions, and the assumption checks with their results. You can run it yourself and get the same numbers.

It is fast because writing the correct script quickly is a tractable problem, and you can trust it because the script is what you receive.

● Four questionsIncluding one it refuses to answer

Watch it show its work.

These are real question shapes from real projects, running against a synthetic cohort. Pick one, or let them play. Switch the output between R and Python and the analysis stays the same.

statlab
You

STATLAB · plan

analysis.RR
Every run ships with the dataset version, the seed, the package manifest and the assumption checks. Nothing on this screen is hidden from you. Escalate to a statistician →
● Under the hoodOur models or NVIDIA's

Choose the model. Keep the code.

STATLAB runs on our own models or on NVIDIA Nemotron, and you choose which. Whichever one writes the analysis, what reaches you is code you can read and re-run.

Our models
Viceron's own models, running inside your perimeter with the rest of STATLAB.
NVIDIA's model
NVIDIA-Nemotron-3.5-Super-VL-120B-A12B, NVIDIA's open-weight model for reasoning, coding and tool use. It runs inside your perimeter too.
Runtime
STATLAB's runtime includes NVIDIA OpenShell, NVIDIA's open-source runtime for AI agents. The agent works inside its sandbox, where policy sets which files it can read or write and where it can connect, and every other outbound connection is blocked.
What you get back
The same with either model: the result, the method paragraph and the R or Python that produced it, seeded and versioned.
● How an analysis runsMeasured in days

The plan comes before the numbers.

Step 01

The question

In your words. The platform turns it into an estimand and shows you the translation before running anything.

  • Population, exposure, outcome
  • Design features detected
  • You confirm or correct
Step 02

Analysis plan

Method, covariates, handling of missing data and the pre-specified sensitivity analyses — written down before the first model.

  • Reviewable document
  • Pre-registerable
  • Deviations logged later
Step 03

Run

Against a frozen dataset version, with a fixed seed, in a pinned environment. Minutes for most questions.

  • Diagnostics run alongside
  • Figures to journal spec
  • Everything logged
Step 04

Verify

You and your statistician both read the code, and where it matters, ours re-runs it independently before you cite it.

  • Independent re-run available
  • Assumption checks reported
  • Alternative specifications
Step 05

Publish

Method paragraph, tables, figures, reporting-guideline checklist and a supplement a reviewer can re-run.

  • Container plus synthetic data
  • Rebuttal support
  • Numbers traceable to a line
● Next doorWhat feeds this

An analysis is only as good as its dataset.

ServiceREDCap Where the frozen, versioned snapshot comes from in the first place. Open → ServiceAI & big data Prediction models, clinical text and imaging at registry scale. Open → ServiceClinical trials Pre-specified analysis plans, interim looks and the DSMB. Open → ServiceSecure data handling Where the model runs, what it sees, and who signs off on it. Open →
Interested? Take the next step

Bring us your impossible.