ProofPrint

ProofPrint is a prototype AI-powered academic tool that audits text segments, flags unverified claims, and cross-references citations against official published DOIs via Semantic Scholar and CrossRef.

Comment

What the Project Does

ProofPrint is a prototype application that helps writers catch fabricated or unsupported claims before they end up in a final paper. It relies on Semantic Scholar and CrossRef to verify bibliographies and citations, ensure they're used properly and make sense, as well as to suggest new sources (with a new citation given) to help better prove the claims the user is making in their writing.

  • Document Ingestion - Users can paste text directly or upload .pdf, .docx, .txt, .md, and .csv files.
  • DOI & Academic Auditing - The system extracts any citations (in-text or bibliography), focusing on officially published DOI sources, and runs parallel database queries against Semantic Scholar and CrossRef to confirm each source actually exists AND says what it's cited as saying.
  • AI Evaluation & Formatting - An AI layer evaluates the factuality of claims and their sources, then formats correction suggestions into APA, MLA, Chicago, or Harvard style per the user's choice.
  • Responsive Reporting - Results are delivered in an Integrity Report with per-claim risk scoring (e.g., 0–100% verified), and the full report can be exported as .txt, .json, or .pdf.

Note: ProofPrint is a prototype and is not completely production-ready, built to demonstrate the concept and workflow that I plan to improve on later. The biggest issue currently is trying to match speed with fact; because of my limited access to stronger programs, I'm unable to allow for any source on the internet to be checked and instead have to rely on Semantic Scholar and CrossRef for all of my information, which means that sources that are just as factual but not part of DOI cannot be considered in ProofPrint at this time.

The Problem Your Project Solves

  • Incorrect/Badly Used Citations - ProofPrint mainly exists to check how well your citations are used; no matter how well the formatting and how plausible the research can seem, ProofPrint can help find ways to improve your citations to make sure that they correctly illustrate the claim you're making. On that end, ProofPrint also verifies whether the claims you are making are actually true, helping ensure the legitimacy of the user's writing.
  • AI Citation Hallucinations - Generative AI models can produce citations that sound completely plausible. They may have correct-looking author names, journals, and years, but if they don't actually exist or don't say what they're claimed to say, ProofPrint will catch these issues before a user submits their work.
  • Prototype Limitations - ProofPrint explicitly deals with the challenge of missing database records: sources that aren't indexed or lack a DOI are graded as "unverifiable" rather than falsely flagged as fake. This avoids penalizing legitimate sources just because they're hard to find in the databases that this application relies on.

Tools Used

Frontend

  • Vanilla JavaScript + HTML5 for frontend & backend design
  • Tailwind CSS for styling (dark violet theme)
  • Fonts: Space Grotesk (display), Inter (body), JetBrains Mono (monospace/code)

File Processing

  • PDF.js - extracts text from uploaded .pdf files in-browser
  • Mammoth.js - extracts text from uploaded .docx files in-browser
  • jsPDF - generates the exportable .pdf Integrity Report

Backend / API

  • Node.js serverless function
  • Featherless AI - LLM inference provider (using Qwen2.5-7B-Instruct as the main model) for all claim evaluation and citation auditing, using JSON-mode prompting
  • Custom claim-extraction pipeline: regex-based sentence parsing, citation-pattern detection (Author, YEAR), and topic-keyword extraction for search queries
  • Concurrency-limited processing to stay within Vercel's serverless timeout (60s max duration)

External Verification APIs

  • Semantic Scholar Graph API - topic-based paper search
  • CrossRef API - bibliographic/DOI-based exact citation lookup
  • Dual-database cross-referencing so a source missing from one index can still be confirmed via the other, reducing false flags

Deployed on Vercel

I (Sage L.) am a solo builder.

Built With

Share this project:

Updates

Submission history