Ship papers withless frictionand a calmer flow.

**Local demo** (no AI calls) Drafted a clean scaffold from your intent. **What this simulates** - Chat-driven drafting and editing - Reference scaffolding (citations / structure) - Real-time LaTeX output you can copy into the editor **Your prompt** > Draft a 1-page research note on contrastive learning with a short abstract, 2 sections, and one equation. ```latex \documentclass[11pt]{article} \usepackage{amsmath,amssymb} \title{A Minimal Research Note on Contrastive Learning} \author{Anonymous} \date{} \begin{document} \maketitle \begin{abstract} Contrastive learning trains representations by pulling related views together while pushing unrelated samples apart. This note sketches a compact intuition and a clean LaTeX scaffold for quick iteration. \end{abstract} \section{Idea} Given two augmented views of the same sample, we maximize agreement in embedding space. \section{Objective} Let $z_i$ and $z_j$ be embeddings of a positive pair and $\tau$ a temperature. The InfoNCE loss is: \[ \mathcal{L} = -\log\frac{\exp(\mathrm{sim}(z_i,z_j)/\tau)}{\sum\limits_{k}\exp(\mathrm{sim}(z_i,z_k)/\tau)} \] \end{document} % Prompt (local demo): % Draft a 1-page research note on contrastive learning with a short abstract, 2 sections, and one equation. ```