Gauntlet-Incept MCP

# Gauntlet Incept Project: Educational Content Generation ## Project Overview The Gauntlet Incept Project aims to build a system that generates high-quality educational content tailored to students' knowledge levels and interests. The initial scope focuses on developing K-8 educational content in the form of articles and question banks for specific subject areas, with a goal to deliver superior learning outcomes. ## Background Incept is being developed as a product that will generate personalized educational content for students. The vision is that this system will eventually serve 1.5 billion students globally, providing tailored educational materials across various subjects, grade levels, and content types. The system will operate using both cloud-based LLMs and device-based SLMs. ### Key Principles - Content will be based on learning science and direct instruction models - Multiple LLMs will be used to generate content for different subjects - Content must be politically neutral and educationally sound - System will leverage Common Core Crawl, a database of high-quality educational content ## Current Project Scope ### Focus Areas - **Grade Levels**: K-8, with current emphasis on: - Language (grades 3-8) - Science (grades 6-8) - Math (grades 6-8) ### Content Types - Direct Instruction (DI) style articles with worked examples - Difficulty-tiered question banks that meet or exceed standardized test requirements ## Assignment Requirements ### Input Resources 1. **Course Definitions** - Subject, grade level, and sequence of lessons - Available at: [Google Sheets Link](https://docs.google.com/spreadsheets/d/1GMCEXqSVFZ-l-nMDGvPl6eOWhuZw29z6rYDhoQgJMZs/edit?gid=0#gid=0) 2. **Common Core Crawl (CCC) Database** - Contains example questions and articles tagged with: - Source (e.g., CK12, Khan) - Content type (question, article, video) - Subject (math, English, science, social studies) - Grade (K-12) - Standard (CCSS, NGSS) - Difficulty level (for questions) - Inventory: [Google Sheets Link](https://docs.google.com/spreadsheets/d/1lOorlDtToCKNcOVBIcLFMYrz3dANSCAWAQAISUqBZl0/edit?gid=0#gid=0) - API Guide: [Google Docs Link](https://docs.google.com/document/d/1CurvQyCGHL6_zuWXeY8--d9dYi0nIse9HSKSatk3yhw/edit?tab=t.0#heading=h.9kn1guva9bzy) - Browser App: [CCC Light Consulting](https://ccc-light-consulting.vercel.app/) 3. **1EdTech Extended QTI Implementation** - Framework for storing generated content - Documentation: [Google Docs Link](https://docs.google.com/document/d/16cIsRjdXXcxOKUXQNzpQ0P86RJk1u9h_AcwXS8IvXIY/edit?tab=t.0) ### Required Outputs 1. **Complete Generated Course** - Display link for the course in a simple visualizer - Course content mapped to QTI structure: - Course = AssessmentTest - Lesson = TestPart - Article Section = Section with worked examples - Article = AssessmentStimulus - Question Bank Section = Section with questions of varying difficulty 2. **Six Functional API Endpoints with Documentation** **For Question Content:** - `tagQuestion`: Identifies subject, grade, standard, lesson, and difficulty of a given question - `gradeQuestion`: Evaluates a tagged question against quality standards, providing pass/fail results with feedback - `generateQuestion`: Creates questions based on specified tags or generates variants of example questions **For Article Content:** - `tagArticle`: Identifies subject, grade, standard, and lesson of a given article - `gradeArticle`: Evaluates a tagged article against quality standards, providing pass/fail results with feedback - `generateArticle`: Creates articles based on specified tags or generates variants of example articles ## Quality Standards ### Question Quality Criteria - Consistent with teaching articles - Appropriate categorization (subject, grade, standard, lesson, difficulty) - Complete with all required components: - Prompt - Interaction type (MCQ, FRQ) - Choices (for MCQ) - Correct answer - Wrong answer explanations - Step-by-step solution - Full explanation - Grading criteria (for FRQ) - Accurate correct answer - No distractors that could be considered correct - At least two plausible distractors for MCQs - No obvious "tells" that point to the correct answer - Clear explanations for wrong answers - Clear solution path - Grade-appropriate language - Consistent, clear, and unambiguous wording - Grammatically correct - Properly formatted ### Article Quality Criteria - Appropriate categorization - Direct Instruction style with worked examples - Step-by-step breakdowns for students with lower working memory - Factually accurate - Grade-appropriate language - Clear and unambiguous wording - Properly formatted - Consistent explanations throughout the course ### Question Bank Quality Criteria - Appropriate categorization - Comprehensive coverage of lesson content - Sufficient depth (at least 100 distinct questions per difficulty level) ## Recommended Implementation Approach 1. **QC-First Architecture** - Build test harness and quality control system before generators - Achieve 99% precision in content quality assessment - Start with one lesson and one difficulty level before expanding 2. **Component Development Sequence** - Develop question generator system first - Then article generator system - Finally course generator system 3. **Key System Components** - Lesson tagger: Maps questions to specific lessons - Test harness: Measures accuracy of quality assessment - Grader: Implements quality standards - Generator: Creates content that meets quality standards - Tagger: Identifies attributes of content ## Success Metrics - Primary: 99%+ precision in quality control - Secondary: High recall (accepting good content) - Focus on preventing low-quality content from reaching students ## Technical Implementation Notes - Use bad examples and mutation testing to improve quality control - Add any low-quality generator output to test harness as bad examples - Maintain a robust regression test suite