SciSpace Academy
TutorialLiterature review › End-to-end research workflow with AI agents

End-to-End Agent Research Workflow: Lessons from a Published Study

Marija ĐukićEnglishLiterature reviewEnd-to-end research workflow with AI agents

TL;DR

1-min read

In this tutorial, Marija Đukić of the University of Belgrade, Faculty of Organizational Sciences, and lead author of a FedCSIS 2026 paper on human-AI collaboration in maturity model development, walks through the real workflow behind her study of Big Data Analytics maturity models. The study drew on dozens of systematic literature reviews and produced a 121-criterion assessment hierarchy. It suits PhD researchers, academics and professionals doing systematic reviews, framework development, model validation or evidence synthesis.

Key takeaways

  • SciSpace can help turn a large body of literature into a structured framework, with the AI proposing a draft that the researcher then tests and refines.
  • Running several rounds of AI-assisted evaluation on your own materials and criteria lets a framework improve through iteration rather than a single pass.
  • Asking the AI to justify each recommendation with sources reduces hallucinations and makes checking faster, provided you open those sources and confirm them.
  • AI adds the most value in knowledge synthesis and in keeping concepts and terminology consistent across a complex hierarchy.
  • AI falls short on practical relevance and methodological judgment, so those decisions should stay with researchers who have domain expertise.
Apply it

To apply this, load your own literature and criteria into SciSpace, request source-backed recommendations, and plan for several review rounds in which you verify and refine the structure yourself.

How to do it

  1. Gather a large body of literature

    Collect the systematic literature reviews and studies relevant to your framework in SciSpace. A broad evidence base gives your end-to-end agent research workflow something solid to structure.

  2. Shape the literature into a framework

    Use SciSpace to organise the evidence into a structured, usable framework, such as a hierarchy of assessment criteria. Structure makes the findings easier to evaluate and refine.

  3. Run iterative AI-assisted evaluation cycles

    Run repeated AI-assisted evaluation rounds on your own materials and criteria, refining the framework after each pass. Iteration surfaces inconsistencies a single pass would miss.

  4. Ask AI to cite each recommendation

    Prompt the AI to back every recommendation with literature. Source-backed answers reduce hallucinations and make checking faster.

  5. Decide where your judgment is required

    Use AI for knowledge synthesis and conceptual consistency, but judge practical relevance and methodology yourself. This keeps the final framework credible.

Watch the tutorial43 min

Building a Research Framework With AI: Lessons From a Published StudyPlay video: Building a Research Framework With AI: Lessons From a Published Study

Chapters

  1. Welcome: what can SciSpace do across a research project?
  2. What is a big data analytics maturity model?
  3. Why repeat a published systematic review with AI?
  4. How do you set the search string and inclusion criteria?
  5. How do you design a data extraction framework?
  6. Which prompt makes AI propose a literature-grounded model?
  7. How do you get AI to critically review its own framework?
  8. How do you run a systematic literature review in SciSpace?
  9. How does SciSpace generate a PRISMA diagram?
  10. How do you extract structured data from many papers with AI?
  11. How does AI flag papers where only the abstract was analysed?
  12. Can AI propose a new research model from the literature?
  13. How do you make AI justify every change with published sources?
  14. Refining the model over five literature-grounded iterations
  15. What changed between the initial and final framework?
  16. Conclusions: AI as a research reviewer, not a replacement

Try this on SciSpace

Frequently asked questions

How can AI help build a conceptual framework from the literature?

AI can scan and summarise many sources, group recurring concepts and propose a draft structure such as a hierarchy of criteria. The researcher then tests that draft against the evidence, removes overlaps and refines it over several rounds. Treating the AI output as a starting point for iteration, not a final answer, keeps the framework defensible.

How do you reduce AI hallucinations in a literature review?

Ask the tool to support every claim or recommendation with specific sources, then open those sources and confirm they say what is claimed. Requiring citations makes unsupported statements easier to spot and speeds up verification. Keep a record of what was checked, and never cite a paper you have not read yourself.

Where does AI fall short in framework or model development?

AI is strong at synthesising existing knowledge and keeping terminology consistent, but it is weaker at judging whether criteria are practical in real settings or whether a method suits a specific context. Those decisions depend on domain expertise and methodological reasoning, so they should stay with the research team.

Related tutorials