Show HN: AI for Science – Curated Resources (github.com)

A new "AI for Science" curated repository compiles hundreds of tools, libraries, papers, datasets and frameworks aimed at accelerating scientific discovery across disciplines—from chemistry and biology to climate and materials. The list groups resources into practical categories (PDF parsing, chart understanding, paper→code, research workbenches, agents, scientific ML, datasets, benchmarks and communities) and highlights state-of-the-art projects such as MinerU and Docling for PDF→structured output, ChartCoder and mPLUG-PaperOwl for chart understanding, Paper2Poster/Auto-Slides/Paper2Slides for automated presentation generation, agentic systems like AI Scientist v1/v2, ChemCrow and Coscientist, and core scientific ML toolkits (PINNs, neural ODEs, DeepXDE, diffrax, Fourier Neural Operator). Technically significant because it stitches the full research pipeline: high-quality document parsing and RAG-ready outputs enable reproducible code generation (AutoP2C, ResearchCodeAgent, ToolMaker); multimodal models and chart-to-code systems let LLMs reason about figures and produce runnable analyses; and numerous benchmarks (ScienceAgentBench, SciTrust, Scientist-Bench) target evaluation of agentic and domain LLMs. The collection lowers adoption barriers for researchers, fosters reproducibility and automated workflows, and surfaces urgent needs—robust benchmarking, trustworthiness checks, and tooling for safe lab automation—as AI becomes an increasingly autonomous partner in scientific discovery.