federicodeponte/opendraft

What it solves

OpenDraft addresses the problem of AI-generated research drafts that contain hallucinated or unverifiable citations. It provides a way to generate long-form academic documents (up to 80+ pages) that are grounded in real scholarly literature and verified against multiple academic databases.

How it works

The system uses a pipeline of 19 specialized AI agents divided into six phases:

1. Research Phase: Finds candidate papers via CrossRef, OpenAlex, and Semantic Scholar, and verifies DOIs against at least two of these sources by default.
2. Structure Phase: Creates a hierarchical research outline with chapters.
3. Writing Phase: Drafts sections using an academic tone.
4. Citation Phase: Dedupes and filters citations, verifying they are relevant to the paper topic.
5. Polish Phase: Refines language and formatting.
6. Export Phase: Generates the final document in PDF, Word, or LaTeX formats.

Who it’s for

• Academic Researchers: Those preparing literature reviews, journal submissions, or thesis first drafts.
• Graduate Students: Students working on master's or PhD dissertations.
• Developers: Those looking to extend a reproducible research-agent pipeline for custom workflows.

Highlights

• Multi-source Citation Verification: Drops citations that cannot be confirmed by at least two scholarly databases (CrossRef, OpenAlex, Semantic Scholar).
• Long-form Output: Capable of generating drafts between 10,000 and 20,000+ words.
• Diverse Export Options: Supports PDF, DOCX, and LaTeX.
• Multilingual Support: Works in over 57 languages.
• Additional Research Tools: Includes a TL;DR 5-bullet summary tool and an audio digest generator using ElevenLabs TTS.