The Document Editor with a
Logic Validation Engine

AI writes fast. Then breaks its own logic. We catch it.

Request Early Access → Watch the Demo
Opens an email to the founders · ~200 Columbia pilot seats open for Q2 2026
Columbia SEAS '28, Austin Wijaya, Akito Yamauchi, Hayden Leung
Real-Time Logic Linting Knowledge Graph AI Verification
01, The Problem

AI-assisted research is everywhere.
So are its artifacts.

In February 2024, a peer-reviewed Frontiers paper on stem-cell biology made it to print with Midjourney-generated figures, including a rat with anatomically impossible anatomy and gibberish labels, and survived peer review for three days before retraction.4 Authors disclosed the AI use. A reviewer flagged concerns. No tool caught it.

768
AI artifacts in print
Published papers flagged for chatbot tells like “regenerate response”, likely a small fraction of the true total.14
19.9%
Fabricated citations
When researchers checked 176 citations GPT-4o generated for mental-health literature reviews, 19.9% were fabricated, the cited papers didn't exist.3
1 of 12
A core GAI risk
NIST, the US federal standards body, names “confabulation” as one of 12 core risks in its 2024 Generative AI Profile. The exact failure mode we address.7

[14] Glynn, Suspected Undeclared Use of AI in the Academic Literature, arXiv:2411.15218, Nov 2024. N=768 cases. · [3] Linardon et al., JMIR Mental Health 2025, N=176 citations. · [4] Frontiers retraction, 13–16 Feb 2024. · [7] NIST AI 600-1, Jul 2024.

02, The Solution

Your editor, with a Logic Validation Engine

We didn't build another AI writer. We built the thing that checks it.

1

Import & Write

Upload PDFs, paste URLs, or draft directly. Sources are parsed and semantically chunked.

2

Graph Builds As You Work

Imported PDFs extract automatically. Live drafts build paragraph by paragraph on commit, typed entities, natural-language relationships.

3

Conflicts Flagged Instantly

Red underline for high-confidence conflicts, orange for likely ones. Click for the source, explanation, and full provenance chain.

Input
Type paragraph
→
Extract
Claims extracted
→
Retrieve
Graph queried
→
Compare
Conflicts surfaced
→
Output
Red underline

Watch the full walkthrough

ResearcherX

Deterministic verification
for every document an AI touches.

Built at Columbia. For work that has to be right.

Working Alpha Columbia SEAS '28
Request Early Access →
Sources & Citations
  1. Linardon, J. et al. “Hallucination Rates and Reference Inaccuracies of GPT-4 Omni in Generating Mental Health Literature Reviews.” JMIR Mental Health 12, e80371, 2025. Peer-reviewed, N=176 citations, Deakin University. mental.jmir.org
  2. Guo et al. “Cellular functions of spermatogonial stem cells in relation to JAK/STAT signaling pathway.” Frontiers in Cell and Developmental Biology, published 13 Feb 2024; retracted 16 Feb 2024. Primary journal retraction record. Science Integrity Digest coverage
  3. Li, J.; Raheja, V.; Kumar, D. “ContraDoc: Understanding Self-Contradictions in Documents with LLMs.” Proceedings of NAACL 2024. Peer-reviewed, 891 documents. aclanthology.org
  4. Chen, X.; Lyu, F.; Liu, K.; Zhang, J.; Lv, X.; Xu, J. “Think Wider, Detect Sharper: Reinforcement Learning for Document-Level Self-Contradiction Detection.” Proceedings of EMNLP 2025. Peer-reviewed, 51.1% accuracy on ContraDoc with SFT + GRPO RL on Llama 3.1-8B. aclanthology.org
  5. NIST AI Risk Management Framework, Generative AI Profile (NIST AI 600-1). US NIST, July 2024. Government framework, 12 GAI risks including “Confabulation.” doi.org/10.6028/NIST.AI.600-1
  6. Glynn, A. “Suspected Undeclared Use of Artificial Intelligence in the Academic Literature: An Analysis of the Academ-AI Dataset.” arXiv:2411.15218, Nov 2024. Open dataset, N=768 published papers flagged for chatbot tells like “regenerate response”. arxiv.org/abs/2411.15218