About

AI agents that end 100 years of academic writing pain.

For the past century, writing a serious academic paper has meant the same soul-crushing ritual. Auto Academic Paper ends it: one casual draft goes in — Markdown, PDF, TXT, even a half-finished homework file — and one finished, citation-verified, publication-ready LaTeX paper comes out, written by a team of AI agents in a single click.

A hundred years of pain

  • Reading 50+ PDFs just to write one paragraph.
  • Synthesizing conflicting theories into a coherent argument.
  • Staring at a blank page to structure a novel theory.
  • Battling reference-manager hell.
  • Wrestling with LaTeX until 3 a.m.

Today, that ends.

One click

One casual draft in. One finished, perfectly formatted, citation-verified, TikZ-beautiful LaTeX paper out. No prompt engineering, no copy-paste loops, no "why did the bibliography break again?"

This isn't another AI writing assistant. It's a team of AI agents powered by state-of-the-art LLMs — academic writing's jump from 1925 to 2025 in a single click.

The pipeline

Seven specialist agents, each with one job:

  • Strategist plans the argument.
  • Librarian researches the literature.
  • Thinker writes the draft.
  • Peer Reviewer checks every claim against the evidence.
  • Rewriter synthesizes and sharpens.
  • Editor formats and places the citations.
  • Appraiser states what the paper claims to contribute.

Bring your own keys — OpenAI, Anthropic, Gemini, Grok, Poe. Every citation is verified against real sources, the LaTeX renders in a live browser preview, and every step is logged in the open — no opaque spinners.

See the open-source project on GitHub →

You are the author

AI can make mistakes. Even with a dedicated Peer Reviewer agent and strict anti-hallucination checks, language models are probabilistic, not deterministic. Always verify the citations, equations, and logic yourself. This tool is the engine — you are the driver.

The maker

Auto Academic Paper is built by Keith Li, a practitioner of Agentic Engineering — the discipline (named by Andrej Karpathy, a step beyond "vibe coding") of directing AI agents under close human supervision to build and harden real production software. He has shipped 10+ market-launched products solo in about six months, and founded Allytic AI to train teams to build the same way.

This is one of his open-source creations. Another is cinematic-3d-battle-engine — a self-playing engine that stages real historical battles in 3D on true-scale satellite terrain, entirely in the browser. Same discipline, a very different extreme.