Free AI detector

Inspiration

As generative AI tools proliferated across university campuses, a quiet crisis emerged in higher education: the false-positive epidemic.

Commercial integrity detectors (such as Turnitin, GPTZero, and Copyleaks) frequently flag authentic student writing as "AI-generated". Because formal academic prose relies on objective syntax, sophisticated vocabulary, and uniform transitional phrases, human-written papers naturally exhibit linguistic properties similar to Large Language Models.

Stressed students were receiving disciplinary notices over coursework written completely from scratch, while commercial services popped up charging \$20 to \$40 per month just to let students check their own work. We built the Academic Wizard Free AI Detector to give every student and researcher a 100% free, transparent, and non-punitive tool to inspect their text's linguistic predictability before submitting.

What it does

The Academic Wizard Linguistic AI Detector analyzes academic drafts in real-time to evaluate how closely sentence structures match algorithmic generation patterns:

• Sentence-Level Visual Heatmap: Rather than returning an opaque percentage score, the tool highlights individual sentences in color-coded tiers based on syntactic predictability.
• Perplexity Scoring: Measures word-choice predictability in context using token entropy. Mathematically, the perplexity ( \mathcal{P} ) of a text sequence ( W = (w_1, w_2, \dots, w_N) ) is evaluated as: $$ \mathcal{P}(W) = \exp\left( -\frac{1}{N} \sum_{i=1}^N \ln P(w_i \mid w_1, \dots, w_{i-1}) \right) $$
• Burstiness Quantification: Evaluates the rhythmic variation in sentence lengths and structure. While LLMs produce uniform sentence lengths, human writing produces natural "bursts" of short and long clauses: $$ \mathcal{B} = \frac{\sigma - \mu}{\sigma + \mu} $$ where ( \mu ) represents the mean sentence length and ( \sigma ) denotes the standard deviation across all sentences in the essay.
• Strict Non-Repository Guarantee: Unlike institutional tools that permanently store student papers in proprietary databases, this utility features a zero-retention policy. Text is processed ephemerally and never saved.
• 100% Free & Unlimited: No credit card, no account required, and no daily quota.

How I built it

• Modern Web Stack: Built with React, Vite, and Tailwind CSS for a lightweight, accessible student-first user interface.
• Linguistic Analysis Engine: Developed client-side tokenizer algorithms evaluating sentence burstiness and transitional cadence.

// Sentence length burstiness and variance calculation
function calculateBurstiness(sentences) {
  if (sentences.length < 2) return 0;
  const lengths = sentences.map(s => s.trim().split(/\s+/).length);
  const mean = lengths.reduce((a, b) => a + b, 0) / lengths.length;
  const variance = lengths.reduce((acc, len) => acc + Math.pow(len - mean, 2), 0) / lengths.length;
  const stdDev = Math.sqrt(variance);
  return (stdDev - mean) / (stdDev + mean);
}

• High-Throughput Parsing: Designed asynchronous parsing loops to process long-form academic papers (up to 10,000+ words) in sub-second time without freezing browser memory.
• Privacy-First Architecture: Text remains strictly in the user's browser session with zero data harvesting.

Challenges I ran into

• The Academic Tone Dilemma: The biggest hurdle was differentiating between synthetic LLM text and rigorous academic prose. Published academic literature is naturally objective, formal, and formulaic. Calibrating the burstiness and perplexity thresholds so that scholarly essays in APA, Harvard, or IEEE styles are not penalized required extensive testing against actual student papers.
• Handling Massive Essay Payloads: Students often inspect full literature reviews and dissertation chapters. Chunking text, mapping sentence coordinates, and updating the DOM heatmap without lag required strict memoization and asynchronous processing.
• Addressing Anxiety Over Detection: Educating students that an AI score is an indicator of syntactic predictability—not an infallible verdict of academic dishonesty.

Accomplishments that I'm proud of

• True Free Access: Delivered a genuine \$0 utility with no hidden paywalls or subscription funnels.
• Actionable Editorial Feedback: Showing students which sentences trigger predictability flags so they can inject authentic authorial cadence and voice.
• Zero Intellectual Property Risk: Ensuring student coursework and research drafts remain 100% private and unindexed.

What I learned

• The mechanics of statistical linguistics and how human prose naturally alternates between concise statements and complex compound clauses.
• The reality of student anxiety regarding automated university integrity software.
• Performance optimization patterns for real-time text analysis in the browser.

What's next for Free AI detector

• Citation-Aware Pre-Filtering: Automatically identifying and excluding APA 7th, Harvard, and OSCOLA in-text citations, block quotations, and reference lists so bibliographies do not skew the score.
• AI Text Cadence Humanizer Integration: A 1-click companion that suggests natural syntactic restructuring for sentences flagged as overly mechanical.
• Browser Extension: A lightweight extension allowing students to run integrity checks directly inside Google Docs and Microsoft Word Online.
• Multilingual Calibration: Expanding detection models to analyze essays written in German, French, and Spanish.

Built With

• python