chatbot-rag-article

# Chatbot RAG Article ## Required Context When working in `chatbot-rag/`, read: 1. `AGENTS.md` 2. `NOTAS_REVISAO.md` 3. `main.tex` 4. Affected files in `sections/` 5. `chatbot-rag.bib` when citations or references are involved 6. Open GitHub Issues for `chatbot-rag` when continuing the editorial backlog Use this skill together with `academic-latex-review` for manuscript-quality work and with `article-action-plan` when executing or updating the GitHub issue backlog. ## Manuscript Positioning - Treat the manuscript as a reproducible proof of concept with potential journal evolution. - Preserve the distinction between preliminary evidence, demonstrated contribution, and open questions. - Position the contribution as a reproducible protocol for Portuguese legislative collections only when the text and evidence support that claim. - Distinguish what is specific to Brazilian Senate speeches from what may transfer to other legislative corpora. - Treat the RAG system as a sociotechnical solution, not merely a technical integration. ## Methodological Priorities - Keep objectives, success criteria, results, and conclusion aligned. - Distinguish retrieval evaluation, generation evaluation, and human evaluation. - Do not present LLM-judge averages as standalone proof of quality. - Relate numerical scores to qualitative examples, variance across runs, rubrics, and human review. - Treat divergence between LLM judge and human reading as a potentially important methodological finding. - For journal-oriented work, prioritize target journal choice, comparative table, Design Science Research framing, propositions, descriptive statistics, visualization, and integrity declarations before large new experiments. - Preserve the current distinction between delivery-ready proof of concept and a stronger journal-submission version that may require new experiments. - Before adding references on RAG, legal AI, legislative corpora, or evaluation frameworks, check `chatbot-rag.bib` for existing related entries and avoid duplicating keys. ## Files and Verification - Main file: `main.tex`. - Sections: `sections/`. - Bibliography: `chatbot-rag.bib`. - PDF output: `out/main.pdf`. - Compile from `chatbot-rag/` with `latexmk -pdf main.tex` after LaTeX edits that affect the PDF. - Check logs for undefined citations, broken references, table issues, and warnings that affect quality. ## Plan Updates - Use GitHub Issues as the operational backlog for future editorial work. - Treat `NOTAS_REVISAO.md` as the consolidated local context for review history, manuscript state, editorial decisions, and the strategic map inherited from the former action plan. - Record progress notes in the corresponding issue with files changed, verification, limitations, and next steps. - Update `NOTAS_REVISAO.md` when the manuscript's overall state changes. - If an action is deferred because it needs new experiments, external data, journal selection, or human annotation, record the deferral as a decision rather than silently skipping it.