Research CV
Identification
Fábio André Malko
Doctoral candidate in Law at IDP (Instituto Brasileiro de Ensino, Desenvolvimento e Pesquisa), Brazil, since 2026. Master of Laws, IDP, 2026. Attorney and accountant. University lecturer.
Undergraduate degrees with institution and year; doctoral programme and research line as recorded at enrolment; teaching institution and courses. None of these should be filled in by inference.
Persistent identifiers
| ORCID | 0009-0006-5975-9603 |
| Lattes | 6791025189749226 |
Education
PhD in Law — IDP, Brasília. In progress since 2026.
Master of Laws — IDP, Brasília, 2026. Dissertation: Automated extraction and semantic treatment of judicial decisions with artificial intelligence: a replicable method applied to research on tax precedents of the Superior Court of Justice (original in Portuguese). Supervisor: Prof. Dr. Luís Felipe Perdigão de Castro.
Output
Dissertation
MALKO, Fábio André. Extração automatizada e tratamento semântico de decisões judiciais com inteligência artificial: um método replicável aplicado a pesquisa de precedentes tributários do STJ. 2026. 198 p. Master’s dissertation (Law) — Instituto Brasileiro de Ensino, Desenvolvimento e Pesquisa, Brasília, 2026. Available at: https://repositorio.idp.edu.br/handle/123456789/5980.
The full text is deposited in open access in IDP’s institutional repository, at the permanent address above, and is available as a PDF file.
The abstract below is the one recorded in the institutional repository.
This dissertation proposes, describes and tests a replicable empirical-computational method for the extraction, structuring and semantic analysis of judicial decisions with the support of artificial intelligence, using state and municipal tax litigation before the Brazilian Superior Court of Justice (STJ) as a testing field. The research question unfolds on two levels: substantively, it asks whether lower courts uniformly apply the precedents of the Superior Court of Justice regarding ICMS, ISS, ITBI and ITCMD; methodologically, which is the core concern, it asks how to process large volumes of decisions in a reconstructible way that is accessible to legal scholars, without prior programming skills and without naive automated reading. The corpus was built from the Court’s own open full-text data, collected through Python web scraping and delimited by the subject codes of the National Council of Justice’s Unified Procedural Table, covering the period from May 2022 to December 2025. The processing was organized into three layers: deterministic extraction, semantic analysis (classification by a language model under a closed category dictionary) and process-level consolidation. Safeguards against hallucination and instability were adopted, along with human validation by sampling. More than fifty-eight thousand full-text decisions were gathered, spread over tens of thousands of unique cases. The findings point to the predominance of admissibility barriers in access to the merits and to a relevant variation in reversal rates among lower courts, consistent with the hypothesis of unevenly distributed divergence, here called islands of resistance. The outcome, proper to a professional master’s degree, is a replicable roadmap for analysing decisions at scale, transferable to other taxes, courts and fields of law.
In preparation
An article on functional replicability in empirical legal research assisted by language models, mapping available instruments and proposing a verification protocol.
Citing this site
MALKO, Fábio André. Replicability in AI-Assisted Legal Research. Available at: https://fabiomalko.org. Accessed: [date].
Academic contact
This address is for academic correspondence. Requests for professional services or legal advice are neither received nor answered through this site.