Crafter Research / Lima, Peru

Public-interest AI research for Peru and LATAM.

We turn public data, legal and civic corpora, and model evaluations into reproducible artifacts, datasets, registries, and accountability tools.

public sourcedatasetevalregistrytooldocs
Team overview

Research should ship artifacts people can inspect.

We work on source-backed datasets, scorecards, registries, evaluation artifacts, and small tools tied to public-interest questions.

data Public data

Messy government and civic sources become reusable, versioned, and inspectable datasets.

corpora Legal and civic corpora

Fragmented public material becomes structured source trails, corpus cards, and research inputs.

evals Model evaluations

Evaluation work should separate correctness, abstention, hallucination, and evidence quality.

Operating model

From source material to reusable artifacts.

Question, sources, artifact, review, documentation, release.

  1. Question

    A public-interest question tied to a real source, corpus, or model behavior.

  2. Sources

    A source trail with URLs, dates, limitations, and evidence quality notes.

  3. Artifact

    A dataset, eval, registry, scorecard, CLI, or accountability tool.

  4. Review

    Check reproducibility, source quality, and limitations before claims harden.

  5. Documentation

    Shareable notes that do not require private vault access to understand the work.

  6. Next gate

    Promote, hold, or reshape the project based on evidence rather than excitement.

Work surfaces

The lab is organized around four artifact types.

Projects can move at different speeds, but the output should stay inspectable and reusable.

Public data Datasets and manifests

Public sources made easier to cite, refresh, verify, and reuse.

Source trails, hashes, refresh logs
Legal and civic corpora Structured source material

Fragmented law, records, and civic text turned into reusable research inputs.

Corpus cards, coverage notes
Model evaluations Evals and scorecards

Evaluation artifacts that make model behavior and evidence quality easier to inspect.

Rubrics, runs, agreement checks
Accountability tools Registries and small interfaces

Focused tools that help people inspect public systems without hiding the source trail.

Registries, CLIs, dashboards
Lab notes Read all notes

Research logs with artifacts.

Measuring whether our legal search actually works Jun 18, 2026 / legalize-pe

We built a hybrid-RAG assistant over 21,000 Peruvian legal norms. Then we measured it, and the conclusion moved three times as the evaluation got more rigorous.
Next research artifact Reserved slot 2

This space stays open until a note has source trail, method, and reusable output.

Next research artifact Reserved slot 3

This space stays open until a note has source trail, method, and reusable output.

Operating stance

Small, practical, evidence-first.

Artifact-first

Every serious claim should point to a dataset, eval, source card set, registry, run, or public method.

Evidence before claims

Public-facing work should be source-backed, limitation-aware, and reproducible enough to inspect.

Shareable docs

Project context should stand on its own without requiring access to private planning notes.