A unified AI detection dashboard that runs your text through multiple commercial detectors simultaneously: ZeroGPT, Sapling AI, Winston AI, and a custom ML model. It aggregates scores and stores scan history for consistent, multi-perspective analysis.
No single AI detector is reliable enough to stake academic integrity decisions on. ZeroGPT might flag something that Sapling misses. Winston might agree with ZeroGPT but for different reasons. Faculty who want to be thorough end up copying text into 3-4 different websites, comparing results manually, and trying to make sense of contradictory scores. That's assuming they even have subscriptions to multiple services.
The process is slow, inconsistent, and leaves no audit trail. When an integrity case goes to committee, faculty need to show their work, which detectors they used, what scores they got, and whether the results were consistent. Without a unified system, that documentation is a mess of screenshots and browser tabs.
Every AI detector has blind spots. Relying on just one produces false positives and false negatives.
Checking text across multiple detectors means copy-pasting into 3-4 websites and manually comparing contradictory results.
When cases go to integrity committees, faculty have no organized record of which tools they used and what they found.
Each detector reports scores differently (0-1, 0-100, inverted scales). Comparing raw outputs is confusing and error-prone.
Built a Node.js backend that calls multiple detection APIs simultaneously (ZeroGPT, Sapling AI, Winston AI) and normalizes all results to a consistent 0-100% format. Each detector service returns a standardized {detector, status, aiScore, humanScore, raw} object.
Built a separate Python Flask service with pre-computed embeddings for a custom machine learning detector that runs alongside the commercial APIs, providing an additional independent signal.
React frontend displays all detector results side-by-side with individual scores, an aggregated consensus view, and visual indicators for agreement/disagreement between detectors.
Every scan is saved to SQLite (or MySQL) with full results from all detectors, creating an audit trail for academic integrity proceedings.
Let's talk about how a unified AI detection workflow could work for your institution.
Start a ConversationSubmit text once, get results from all enabled detectors in parallel. 30-second timeout per API, 60 seconds for the ML model.
Every detector's output is converted to a consistent 0-100% AI/Human score, regardless of how the original API reports results.
A Python-based detector using pre-computed embeddings that provides an independent, non-commercial detection signal alongside the API results.
Every scan is persistently stored with timestamps, input text, and per-detector results. Searchable history for integrity case documentation.
Health check endpoint monitors API availability. Disabled detectors (Copyleaks, GPTZero) can be enabled when credentials are available.
Sapling integration provides per-sentence AI probability scores, showing which specific sentences are most likely AI-generated.
Side-by-side results from all enabled detectors with normalized scores and an aggregated consensus view.
Persistent scan history with timestamps, text previews, and per-detector score summaries for audit trail documentation.
Sapling's per-sentence AI probability scores highlighting which specific sentences are most suspicious.
Instead of maintaining subscriptions to four different detection tools and manually comparing results, everything is in one place. The scan history has saved us multiple times when integrity cases went to the appeals committee.
ZeroGPT returns a “fakePercentage,” Sapling returns a 0-1 probability, Winston returns a “human score” where higher means more human. Getting all of these onto a consistent scale where results are actually comparable required careful per-service normalization logic.
A document might be 60% AI overall, but that doesn't help faculty. Sapling's per-sentence breakdown shows exactly which paragraphs are suspicious, that's actionable information for a conversation with a student.
The detection scores themselves are useful, but the real value is the persistent, searchable history. Academic integrity cases can take weeks to resolve and go through multiple committees. Having a timestamped record of exactly what was scanned and what each detector found is what faculty actually need.
Tell us about your academic integrity process and let's explore what a multi-detector aggregation system could look like for your institution.
No pitch. No pressure. Just a conversation about what might work.