SlopSift: The Local NLP Linter That Sharpens AI and Human Writing

SlopSift isn't just another grammar checker, it's a surgical tool for tightening prose, whether human or AI-generated. By mapping the relationships between words, it spots the subtle flaws that make writing weak: canned arguments, unsupported claims, and filler text.

The core of SlopSift is a compact NLP model that runs entirely on your device. This isn't a cloud API wrapper; it's a quantized ONNX model that works in Node.js or your browser via WebAssembly. Privacy? Check. Offline use? You got it.

import SlopSift from 'slopsift';

const text = 'As an AI language model, I cannot have personal opinions.';
const results = SlopSift.lint(text);
console.log(results);
// Output: [{ type: 'error', message: 'Caught red-handed: AI boilerplate detected' }]

But SlopSift's real power lies in its rule system. It combines the NLP model's grammatical graph with deterministic checks to find structural tells. Every finding points to the exact text that triggered it, categorized as errors (strong tells), warnings (needs attention), or notes (worth a look).

One killer feature: SlopSift catches repetitive structures. When three paragraphs use the same canned outline, it calls that out. It's not about word-for-word plagiarism; it's about the subtle sameness that makes AI-generated text feel robotic.

SlopSift doesn't pretend to be an AI detector. It doesn't care who typed a sentence. Instead, it focuses on substance: Does this claim have support? Is this passive voice hiding responsibility? Is this just filler?

For developers, SlopSift integrates smoothly into workflows. It can glob files, lint Markdown, inspect code comments, and spit out ESLint-compatible JSON. Run it in CI, or let your coding agents use it to refine their output without flattening your voice.

The tool's development process is a lesson in focused machine learning. Starting with a compact pretrained English encoder, the team fine-tuned it on parts of speech and dependency parsing. They used structured distillation from a larger model, then honed it with 50 carefully crafted examples targeting the grammatical relationships their linter rules would use.

SlopSift's approach raises the bar for writing tools. By focusing on structure rather than origin, it offers something more valuable than yet another dubious AI detector. It's a tool for making all writing, human or machine, more precise and impactful.

Yes, AI helped build SlopSift. That's exactly the point. As AI-generated text becomes ubiquitous, we need tools that improve content quality, not just flag its source. SlopSift shows us what that future might look like: local, privacy-preserving, and focused on making every sentence count.