How we use AI - Lumono

How we operate

Our principles

We built Lumono around three commitments that shaped the architecture before they shaped the marketing.

Transparency

No black boxes. Every step of the Lumono workflow is visible to the user — how the AI interpreted your question, which variables it suggested, the R code it generated, and the analysis output. Nothing happens behind a curtain.

Reproducibility

Every analysis ships with its receipts. Lumono hands you the R code for every analysis you run. Take it to your own machine, share it with a collaborator, or submit it for peer review — the work is portable, auditable, and yours.

Integrity

We designed Lumono to operate within the constraints data custodians set, not to work around them. Where we had a choice between convenience and caution in handling data, we chose caution.

ai in the research workflow

Where AI is in Lumono — and where it isn't

AI handles the parts of research that benefit from a thinking partner — framing a question, navigating a codebook, drafting an analysis plan. It doesn't touch the data, run the statistics, or produce the results.

Where we use AI

Three specific places in the research workflow:

Refining your research question. AI helps you move from a broad interest to a methodologically sound, answerable study question.

Identifying relevant variables. AI reads publicly available data dictionaries and codebooks to suggest which variables map to your question.

Generating an analysis plan. AI drafts an appropriate statistical analysis as R code, tailored to your study design.

Where we don't

The boundaries that protect your data

Handling person-level data. Subject-level records never leave our infrastructure and are never sent to an AI provider.‍

Running statistical analysis. All analysis executes in R inside Lumono's environment. The AI provider doesn't compute results.

Storing your work with third parties. Our AI provider operates under a zero data retention agreement. Nothing about your research is retained by them.

Data boundaries

What we send to LLMs — and what we don't

Our AI use isn't via consumer products like ChatGPT, Claude.ai, or Gemini — it's via enterprise APIs under contractual zero data retention, which is the practice the public-LLM prohibitions in many data use agreements exist to address.

Sent to LLM vendors

Only what's needed to help you design a study

Variable names, descriptions, value labels, counts, proportions, and standard deviations from data dictionaries — the same metadata that already appears in published codebooks for these datasets.

Your research question and study design inputs

Analysis plan structure (population, exposure, outcome, and confounder definitions)

Never sent to LLM vendors

The data that stays inside our infrastructure

Person-level data or subject-level records

Any data covered by a dataset custodian's data use agreement

Proprietary dataset content of any kind

How analysis actually works

Five steps from your research question to reproducible results. AI is present in the first three steps — and absent from the rest.

Datasets live in Lumono's infrastructure. They're never copied, distributed, or sent to external services.

You design your study with AI assistance. Lumono helps you refine your question and identify variables from public codebooks.

Lumono generates an R analysis plan. The plan is visible to you before it runs. You can review, edit, or reject it.

The analysis executes in R, inside Lumono's environment. Not in an AI provider. Not in a vendor environment. The R statistical engine runs the analysis against the dataset.

You receive results and the R code. You can rerun the same analysis on your own infrastructure to verify every result.

For dataset custodians and institutional reviewers

If you're evaluating Lumono for a data use application, an institutional partnership, or a security review, we'd welcome the conversation. We can provide technical documentation, a walkthrough of our data architecture, BAA references, and answers to specific provisions in your AI or data use policy.
‍
Reach us at trust@lumono.ai.