By name: ai-project-x-ray description: X-ray vision for AI/LLM projects - analyzes Streamlit apps, LLM pipelines, and educational platforms to reveal architecture, API dependencies, scoring logic, and potential bottlenecks. Use when you need instant insight into how an AI project works. allowed-tools: Read, Grep, Glob, Bash
AI Project X-Ray
Perform a deep structural analysis of AI/LLM projects to reveal:
1. Project DNA
- Type Classification: Streamlit lesson app / LLM orchestration / Content generator / Testing interface
- Core Purpose: What problem does this solve?
- User Journey: How do users interact with it?
2. LLM Anatomy
- API Dependencies: Which LLM providers (OpenAI, Gemini, Groq, Hugging Face)?
- Model Specifics: Which models? (GPT-4o, Gemini-1.5-pro, Llama-3.3, etc.)
- Prompt Engineering: System prompts, temperature settings, output formats
- Fallback Strategy: What happens when APIs fail?
3. Architecture X-Ray
- Entry Points: Main execution files
- Key Modules: Core logic engines, API clients, scoring systems
- Data Flow: Input → Processing → Output pipeline
- Session Management: State handling (Streamlit session state keys, etc.)
4. Scoring & Validation Logic
- Grading Systems: Heuristic vs LLM-based scoring
- Evaluation Criteria: What makes a "good" response?
- JSON Schemas: Expected output formats
- Error Handling: Validation and fallback patterns
5. Health Check
- Security: API key management, secret handling
- Testing: Test coverage and validation
- Dependencies: Critical packages and versions
- Bottlenecks: Potential performance issues or fragile points
6. Quick Start Recipe
- Installation steps
- Required environment variables
- Run command
- Expected behavior on first launch
Analysis Method
- Check project documentation (CLAUDE.md, README, copilot-instructions)
- Examine entry point (app.py, index.html)
- Read requirements/package files for dependencies
- Search for API client code and prompt engineering
- Analyze scoring/evaluation logic if present
- Review session state and data flow patterns
- Check test coverage and error handling
- Provide actionable insights and improvement suggestions
Output Format
Present findings as a structured report with:
- Clear section headers
- Code references with file:line_number format
- Visual indicators for health status (healthy/warning/critical patterns)
- Actionable recommendations at the end