Crafter Research / Lima, Peru
We build corpora, evals, agents, and working interfaces for the messy institutions people actually depend on.
LATAM gives us brittle portals, fragmented law, weak APIs, and Spanish-language gaps. We turn those constraints into inspectable AI systems.
Government portals, fragmented law, and civic data become research settings.
Every claim should point to a corpus, benchmark, run, interface, or repo.
Language and institutional gaps are treated as first-class evaluation targets.
Question, corpus, eval, agent, interface, research log.
A research question tied to a real civic system.
Public data made reusable, versioned, and citable.
A harness that can make the idea fail in public.
Retrieval or tool-use systems built from the artifacts.
A working surface people can inspect or reuse.
The argument linked back to code, data, and runs.
Legal AI, language evals, gov adapters, and civic data systems.
Open legal corpora, retrieval evals, and citation-grounded assistants for Peruvian law.
Corpus, SDK, paper trackLatin American Spanish evals that separate correctness, abstention, and hallucination.
Harness, judge artifactsAgent-first CLIs that turn hostile public portals into reliable tools.
Adapters, audit logsPublic data products for finance, elections, and institutional relationships.
Maps, graphs, datasetsCivic tech is the field site, not the whole identity.
A working system should make the paper more inspectable, not just more marketable.
Messy public data, weak APIs, and Spanish-language gaps are research constraints worth studying.
