AI evidence extraction

Como Criar seu Próprio Assistente de Pesquisa Científica com LangGraph

Larissa Souza do Nascimento, Ricardo Moura Sekeff Budaruiche · 2025

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

9/10

Relevance

1/4

Quality (LMQS)

Evidence

Citations

0.00

FWCI

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.5753/sbc.16935.3.3

Methodology & findings

Study design

Design science/artifact development with case study implementation.

Primary method

design science research; tutorial-based practical implementation; exploratory system design

Main result

The study demonstrates that "o LangGraph pode ser utilizado para criar agentes capazes de interagir com repositórios científicos, como o arXiv, integrando dados, refinando respostas e auxiliando na geração de conhecimento" (LangGraph can be used to create agents capable of interacting with scientific repositories, such as arXiv, integrating data, refining responses and assisting in knowledge generation). The multiagent architecture with supervisor coordination shows effectiveness for automating bibliographic searches and accelerating access to scientific information.

Research paradigm

pragmatist/design science

Author conclusions

"O impacto esperado da aplicação prática do LangGraph vai além da automação. Espera-se que essa tecnologia contribua para a democratização do acesso à informação científica, otimize processos de revisão de literatura e reduza barreiras técnicas para pesquisadores iniciantes ou com menor familiaridade com ferramentas computacionais" (The expected impact of practical LangGraph application goes beyond automation. It is hoped that this technology will contribute to democratizing access to scientific information, optimize literature review processes and reduce technical barriers for beginning researchers or those less familiar with computational tools). The authors emphasize that "a formação de pesquisadores sobre o uso ético, crítico e transparente da IA será determinante para o sucesso de iniciativas como esta" (training researchers on ethical, critical and transparent use of AI will be decisive for the success of initiatives like this).

Risk of bias

Algorithmic bias in LLMs trained on internet data reflecting social inequalities and stereotypes; Bias in article retrieval and source prioritization by the arXiv and Tavily agents; Opaqueness of Claude 3.5 model affecting transparency of decision-making; Selection bias in scientific articles retrieved from arXiv may favor certain research areas or publication patterns; Bias in source prioritization and article interpretation by agents; Training data bias in LLMs (acknowledged: models trained on internet data reflecting social inequalities and stereotypes); No systematic evaluation or user study conducted; No comparative benchmarking against alternative frameworks; No quantitative performance metrics reported

Limitations

The authors explicitly state: "Durante o desenvolvimento, foram enfrentadas limitações quanto à estabilidade das ferramentas de IA utilizadas, controle de alucinações, necessidade de chaves privadas, além de desafios relacionados à explicabilidade do comportamento dos agentes" (During development, limitations were faced regarding the stability of AI tools used, control of hallucinations, need for private keys, and challenges related to agent behavior explainability)
Additionally, "ainda são necessárias melhorias quanto à curadoria de fontes, integração com bancos de dados institucionais e definição de métricas confiáveis de desempenho" (improvements are still needed regarding source curation, integration with institutional databases and definition of reliable performance metrics).

Open questions raised

Integration with institutional databases (SUAP, IFPI repositories)
User profile personalization (student, teacher, researcher adaptations)
Collaborative evaluation mechanisms by user communities
Multimodal capabilities (images, graphics, videos)
Source curation improvements
Reliable performance metrics definition

Data: arXiv repository (via arXiv API); ChromaDB local vector store for semantic storage of scientific articles; arXiv API (https://arxiv.org/ - referenced for paper retrieval); ChromaDB vector store (local persistence mentioned but no public dataset provided)Code: https://github.com/langchain-ai/langgraph/blob/main/docs/docs/tutorials/multi_agent/agent_supervisor.ipynb; GitHub repository referenced: https://github.com/langchain-ai/langgraph/blob/main/docs/docs/tutorials/multi_agent/agent_supervisor.ipynbExtracted from: pdfAgreement 64%

Explore related topics

RAG for Research Literature Discovery Design Science