Opencanvas – weekend project by mai/Google/Anthropic engineers (github.com)

OpenCanvas is an open-source, AI-powered slide-generation tool (repo: github.com/genmini-ai/OpenCanvas) that converts topic prompts or PDF papers into polished HTML slide decks and high-quality PDFs. It supports dual inputs (text topics or PDFs with optional image extraction), automatic web research (via Brave Search), multiple professional themes, and a full REST API/CLI. Crucially, it integrates automated quality evaluation using Claude, OpenAI GPT, or Gemini and includes an autonomous “evolution” loop that evaluates outputs, identifies improvements, updates prompts/tools, and iteratively regenerates slides to raise presentation quality. Outputs are saved in timestamped folders with source tracking; rendering uses Playwright (Chromium) with a Selenium fallback and configurable PDF zoom. For the AI/ML community this matters because it streamlines the end-to-end workflow from literature or ideas to conference-ready slides and embeds continuous evaluation-driven improvement — speeding preparation for talks, tutorials, and reproducible demos. Technical implications: generation relies on Anthropic Claude (ANTHROPIC_API_KEY required) while evaluation can be routed to other provider APIs; the system is configurable (evaluation model/provider, themes, zoom) and extensible via FastAPI endpoints. Being MIT-licensed and modular (Playwright, FastAPI, Brave Search) makes it easy to adopt or adapt, but users should note API-key dependencies, potential model bias in automated evaluations, and provenance/trust considerations when relying on automated content and web research.