De l’OCR au RAG : l’IA générative appliquée aux sources parlementaires françaises. Enjeux, méthodes et perspectives

Value of the Information & Strength of the Argument

The talk provides valuable insights into the application of generative AI to historical research, particularly parliamentary debates. The speaker demonstrates a solid understanding of both historical methodology and AI technologies. She clearly articulates the research question and the theoretical framework (agenda-setting). The argumentation is coherent, moving from the research context to the technical pipeline and challenges. However, the presentation is more descriptive than analytical, with limited concrete results or case studies. The speaker acknowledges the experimental nature of the project, which adds credibility but also limits the depth of the argumentation.

Scientific Rigor, Source Quality, Title Accuracy

The speaker references several academic works, including those by Frédéric Clavert and Caroline Müller on AI and history, and the agenda-setting theory by McCombs. She also mentions the digital turn concept by Pasi Ihalainen. These references are appropriate and lend credibility. However, the talk does not provide a systematic review of sources, and some claims are made without detailed citations. The title accurately reflects the content, and the presentation is well-structured. The speaker’s expertise is evident, but the lack of empirical data and the early stage of the project limit the scientific rigor.

The title accurately reflects the content, which covers the pipeline from OCR to RAG applied to parliamentary sources.

7/10

The speaker is a historian and researcher at EPITA with expertise in digital humanities. The talk presents a research project in its early stages, with methodological reflections and references to academic works. However, the presentation is largely descriptive and lacks detailed empirical evidence or peer-reviewed results.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 02:00Introduction and presentation of the project 'Décidons'.
• 05:00Context of parliamentary history and digitization efforts.
• 10:00Research question on agenda-setting and media influence.
• 15:00Challenges of using LLMs for historical research.
• 20:00Introduction to RAG and its potential for grounding LLMs.
• 25:00Technical details of the pipeline: OCR, segmentation, vectorization.
• 30:00Discussion on reproducibility and prompt sensitivity.
• 35:00Example of LLM hallucination with GitHub Copilot.
• 40:00Future perspectives and interdisciplinary collaboration.
• 45:00Conclusion and Q&A.

• Clavert, F., & Müller, C. (2024). IA et histoire : de l'impact à l'usage. — Referenced as an article on the impact of AI on the historian's profession.
• Müller, C. (2025). [Book on AI and history] — Referenced as a book by Caroline Müller on AI and history.
• McCombs, M. (1972). The agenda-setting function of mass media. — Referenced as the theory of agenda-setting.
• Ihalainen, P. (2017). The Digital Turn in Parliamentary History. — Referenced as the origin of the 'digital turn' concept.

• Clavert, F., & Müller, C. (2024). IA et histoire : de l'impact à l'usage. — Aligns with the speaker's discussion on AI's impact on history.
• McCombs, M. (1972). The agenda-setting function of mass media. — Supports the theoretical framework.

• No discordant sources identified. — The talk does not present conflicting sources.