← Projects

ResearcherX

An AI writing environment with a logic-validation engine that flags contradictions in AI-assisted drafts, with line-level provenance.

Overview

ResearcherX is an AI document editor. I’m one of its three co-founders and a product lead. It builds a knowledge graph from your own sources and checks AI-assisted drafts against it in real time, so a claim that contradicts your references gets caught as you write.

The problem

AI writing tools are fluent, which is exactly what makes them dangerous for research writing: a draft can state something that quietly contradicts a paper you cited three pages earlier, and it reads perfectly well. Catching that by eye across a long document is precisely the kind of consistency-checking people are bad at and give up on.

How it works

Your sources become a knowledge graph. Retrieval combines that graph with semantic embeddings and full-text search, so the system can find the relevant prior claims even when the wording differs. A logic-validation step then checks the draft against them and surfaces conflicts with line-level provenance — pointing back to the exact source a sentence contradicts, rather than just flagging that something is off. The system is model-agnostic, private by default, and self-hostable, so a lab can run it without sending its unpublished work to anyone.

Traction

We ran a roughly 200-student beta pilot at Columbia. ResearcherX placed 2nd at the Columbia Lion’s Cage and was a finalist in the Millard Chan ’99 Technology Challenge — at the latter, among the only undergraduate teams in the field.