@inproceedings{batitucci-etal-2026-agent,
title = "Agent Orchestration - {LLM} for Legal Metadata Extraction: A Comparative Analysis of Efficiency and Precision",
author = "Batitucci, Luiz An{\'i}sio and
Lopes, Luciane In{\'a}cia and
Ferreira, Rhodie and
Paraiso, Emerson Cabrera",
editor = "Souza, Marlo and
de-Dios-Flores, Iria and
Santos, Diana and
Freitas, Larissa and
Souza, Jackson Wilke da Cruz and
Ribeiro, Eug{\'e}nio",
booktitle = "Proceedings of the 17th International Conference on Computational Processing of {P}ortuguese ({PROPOR} 2026) - Vol. 1",
month = apr,
year = "2026",
address = "Salvador, Brazil",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.propor-1.72/",
pages = "727--737",
ISBN = "979-8-89176-387-6",
abstract = "This work introduces and evaluates JAMEX (Judicial Multi-Agent Metadata Extraction), a multi-agent pipeline for extracting structured metadata from Brazilian court decisions (Espelho do Ac{\'o}rd{\~a}o), and compares it against a strong single-prompt baseline under an Information Retrieval-only (IR-only) setting.We first ran a pilot on 300 decisions and then reran the experiment on a stratified dataset of n=1,225; completion rates varied across executions, yielding between 779{--}1,216 successfully completed instances, with non-completion concentrated in agentic configurations.Across re-executions, the accuracy impact of agents was strategy-dependent: GPT-5 improves over the baseline in multiple agentic strategies but not across all orchestration variants, while smaller models (Gemma3-12B/Gemma3-27B) show no robust gains.Orchestration refinements motivated by agent design literature (memory, planning and directed review) improved traceability, but performance remained sensitive to task decomposition and context splitting.Overall, JAMEX increases token usage and operational complexity, so deployment must balance accuracy, completion reliability, and cost for Portuguese legal metadata extraction."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="batitucci-etal-2026-agent">
<titleInfo>
<title>Agent Orchestration - LLM for Legal Metadata Extraction: A Comparative Analysis of Efficiency and Precision</title>
</titleInfo>
<name type="personal">
<namePart type="given">Luiz</namePart>
<namePart type="given">Anísio</namePart>
<namePart type="family">Batitucci</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Luciane</namePart>
<namePart type="given">Inácia</namePart>
<namePart type="family">Lopes</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rhodie</namePart>
<namePart type="family">Ferreira</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Emerson</namePart>
<namePart type="given">Cabrera</namePart>
<namePart type="family">Paraiso</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-04</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 1</title>
</titleInfo>
<name type="personal">
<namePart type="given">Marlo</namePart>
<namePart type="family">Souza</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Iria</namePart>
<namePart type="family">de-Dios-Flores</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Diana</namePart>
<namePart type="family">Santos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Larissa</namePart>
<namePart type="family">Freitas</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jackson</namePart>
<namePart type="given">Wilke</namePart>
<namePart type="given">da</namePart>
<namePart type="given">Cruz</namePart>
<namePart type="family">Souza</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Eugénio</namePart>
<namePart type="family">Ribeiro</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Salvador, Brazil</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-387-6</identifier>
</relatedItem>
<abstract>This work introduces and evaluates JAMEX (Judicial Multi-Agent Metadata Extraction), a multi-agent pipeline for extracting structured metadata from Brazilian court decisions (Espelho do Acórdão), and compares it against a strong single-prompt baseline under an Information Retrieval-only (IR-only) setting.We first ran a pilot on 300 decisions and then reran the experiment on a stratified dataset of n=1,225; completion rates varied across executions, yielding between 779–1,216 successfully completed instances, with non-completion concentrated in agentic configurations.Across re-executions, the accuracy impact of agents was strategy-dependent: GPT-5 improves over the baseline in multiple agentic strategies but not across all orchestration variants, while smaller models (Gemma3-12B/Gemma3-27B) show no robust gains.Orchestration refinements motivated by agent design literature (memory, planning and directed review) improved traceability, but performance remained sensitive to task decomposition and context splitting.Overall, JAMEX increases token usage and operational complexity, so deployment must balance accuracy, completion reliability, and cost for Portuguese legal metadata extraction.</abstract>
<identifier type="citekey">batitucci-etal-2026-agent</identifier>
<location>
<url>https://aclanthology.org/2026.propor-1.72/</url>
</location>
<part>
<date>2026-04</date>
<extent unit="page">
<start>727</start>
<end>737</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Agent Orchestration - LLM for Legal Metadata Extraction: A Comparative Analysis of Efficiency and Precision
%A Batitucci, Luiz Anísio
%A Lopes, Luciane Inácia
%A Ferreira, Rhodie
%A Paraiso, Emerson Cabrera
%Y Souza, Marlo
%Y de-Dios-Flores, Iria
%Y Santos, Diana
%Y Freitas, Larissa
%Y Souza, Jackson Wilke da Cruz
%Y Ribeiro, Eugénio
%S Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 1
%D 2026
%8 April
%I Association for Computational Linguistics
%C Salvador, Brazil
%@ 979-8-89176-387-6
%F batitucci-etal-2026-agent
%X This work introduces and evaluates JAMEX (Judicial Multi-Agent Metadata Extraction), a multi-agent pipeline for extracting structured metadata from Brazilian court decisions (Espelho do Acórdão), and compares it against a strong single-prompt baseline under an Information Retrieval-only (IR-only) setting.We first ran a pilot on 300 decisions and then reran the experiment on a stratified dataset of n=1,225; completion rates varied across executions, yielding between 779–1,216 successfully completed instances, with non-completion concentrated in agentic configurations.Across re-executions, the accuracy impact of agents was strategy-dependent: GPT-5 improves over the baseline in multiple agentic strategies but not across all orchestration variants, while smaller models (Gemma3-12B/Gemma3-27B) show no robust gains.Orchestration refinements motivated by agent design literature (memory, planning and directed review) improved traceability, but performance remained sensitive to task decomposition and context splitting.Overall, JAMEX increases token usage and operational complexity, so deployment must balance accuracy, completion reliability, and cost for Portuguese legal metadata extraction.
%U https://aclanthology.org/2026.propor-1.72/
%P 727-737
Markdown (Informal)
[Agent Orchestration - LLM for Legal Metadata Extraction: A Comparative Analysis of Efficiency and Precision](https://aclanthology.org/2026.propor-1.72/) (Batitucci et al., PROPOR 2026)
ACL