The Evidence Leashed AI: Using Literature Reviews, Methodological Audits, Lineage Maps, and Contradiction Logs to Train Gemini Gems for Biology Research Proposals

The Pedagogical Workflow This curriculum guides students through a highly scaffolded, four-stage research pipeline: The Methodological Audit: Students analyze a library of 20 high-level papers, extracting fine-grained variables (e.g., sample sizes, model systems, animal models, time points, and statistical tests) rather than just reading abstracts. Lineage Mapping & Contradiction Logs: Students map out how these papers relate to one another and construct a "Contradiction Log" pinpointing exactly where and why the published literature disagrees. Training the Gemini Gem: Students import their manual audit data and contradiction logs into a custom Gemini Gem, programming it with strict instructions to prioritize experimental design flaws over general summaries. Stress-Testing & Proposal Defense: Students use their custom-trained Gem to find the "weakest links" in their own biological reasoning, using the AI's critique to harden and refine their final research proposals. What is Included in This Repository? This project provides all the open-source materials needed to adapt and implement this framework in your own science classroom: Syllabus & Assignment Prompts: Step-by-step student-facing instruction sheets for each phase of the project. Methodological Audit & Contradiction Log Templates: Downloadable spreadsheet templates (Excel/Google Sheets) for student data collection. AI Instruction & System Prompt Guides: The exact system instructions and guardrails used to program the Gemini Gems to act as critical auditors. Grading Rubrics: Clear assessment criteria for evaluating both the manual scientific auditing and the final "hardened" research proposals. Why Adopt This Framework? Inhibits AI Plagiarism: Because the AI is strictly "leashed" to the student's own manual database, students cannot use AI to bypass the critical thinking or reading phases of the project. Teaches Modern Data Literacy: Students gain hands-on experience in scientific data architecture, learning how to structure data to guide AI tools. Highly Replicable: While developed for an immunology capstone, this workflow easily adapts to any upper-level STEM course focused on proposal writing and literature synthesis.