What problem does it solve? Creating figures for ML conference papers requires choosing the right chart type, matching venue-specific dimensions and fonts, and producing architecture diagrams that look professional rather than generic. This Skill automates figure generation from raw research context, results tables, or system descriptions. ## Core Features & Use Cases - Architecture Diagram Generation: Extracts system components and relationships from paper text, then generates styled diagrams (hand-drawn sketch, modern minimal, illustrated technical, or classic academic) via Gemini image generation with structured 6-section prompts. - Data-Driven Charts: Auto-selects chart types (line plots, grouped bars, heatmaps, scatter, leaderboards) from experiment results and renders them with matplotlib/seaborn using colorblind-safe palettes and venue-correct sizing. - Publication Styling: Enforces NeurIPS/ICML/ICLR/ACL/AAAI figure dimensions, Times font matching, PDF vector export, and highlights the paper's own method with a distinct color. - Use Case: Given a results paragraph like "GPT-4: MMLU 86.4, Ours: 88.1", the Skill detects a grouped bar comparison, highlights "Ours" in coral, and exports both PDF and 300-DPI PNG files ready for LaTeX inclusion. ## Quick Start Ask the agent to create a figure for your paper by pasting your method section or experiment results, for example: "Generate an architecture diagram from this system description and a grouped bar chart from these benchmark scores."
