Applied research and evidence

Real questions, reviewable methods, and outcomes with visible limits.

A public applied-research agenda, reproducibility criteria, and connected technical profiles without invented publications or outcomes.

Public agenda

Research to improve a decision, not to decorate a profile.

My research work focuses on applied problems where architecture, evaluation, governance, and product must hold together. This page presents the agenda and standard; it does not turn internal materials, tests, or open profiles into publications.

Research directions

Four questions connected by one evidence discipline.

01

Applied AI systems

How to turn models, data, tools, and human oversight into operational capabilities that can be evaluated.

02

Agent evaluation

How to measure usefulness, boundaries, authority, traceability, and adversarial behavior before expanding autonomy.

03

Reproducible knowledge

How to preserve sources, versions, decisions, and sufficient conditions to repeat or falsify an outcome.

04

Product and evidence

How to connect a technical question to a product decision without turning a demonstration into commercial impact.

Evidence method

From a question to an artifact that can be reviewed.

  1. 1

    Question

    Define what needs to be learned, for whom, and which observation could change the decision.

  2. 2

    Protocol

    Fix data, environment, comparison, metrics, risks, and invalidation conditions before seeing the outcome.

  3. 3

    Artifact

    Version allowed code, data, configuration, output, and enough lineage for inspection.

  4. 4

    Evaluation

    Separate local tests, repetitions, boundaries, and measured outcomes without generalizing beyond the evidence.

  5. 5

    Publication

    Publish only after reviewing identity, rights, sensitivity, reproducibility, and claim language.

Connected record

Identity, reproducible materials, and technical work remain separate.

These public profiles serve different purposes. Their existence confirms an observed surface, not an approved work, dataset, notebook, or outcome.

Research identity

ORCID

Persistent identifier for connecting verifiable works when they exist.

Open ORCID
Reproducible knowledge

OSF

Space for reproducible materials that pass rights and scope review.

Open OSF
Data and notebooks

Kaggle

Technical profile for real notebooks or datasets under clean-room and rights review.

Open Kaggle
Authorized verified publications: 0

The record will contain only real, versioned, reproducible, and approved artifacts. No placeholder content is used.

Publication standard

Four conditions before making an outcome claim.

Source and lineage
The claim points to identifiable sources and versions.
Reproducible method
Another person can understand what ran and under which conditions.
Rights and privacy
Code, data, images, and text have publication-compatible authorization.
Scope and limitations
The outcome states what it shows, what it does not show, and when it must be reviewed.

Collaboration

An applied question can begin with context, not a publication promise.

If you need to evaluate a hypothesis, system, or AI opportunity, we can first define the question, scope, and exit criterion.