Detecting-AI.com is an AI-powered platform that detects whether text is written by a human or generated by artificial intelligence. We also provide tools for plagiarism detection and fact-checking. Our mission is to help educational institutions, publishers, and organizations build trust and transparency in the age of AI.
This repository is created for demonstration and evaluation purposes for the President Tech Award (awards.gov.uz). It includes detailed information about the technology, use cases, and our model architecture, without exposing sensitive or private source code.
📺 Elevator Pitch Video
🎯 Language: English / Uzbek
⏱ Duration: ~2 minutes
- For Educators & Universities: Helps identify AI-written student work and maintain academic integrity.
- For Publishers & Businesses: Detects synthetic content to prevent misinformation and reputation risks.
- For Institutions: Supports internal compliance by ensuring written materials are original and verifiable.
- Over 250,000 registered users
- 10,000+ daily active visitors
- Used by educators and students in the U.S., Brazil, UK, Canada, Uzbekistan, and more
- Fully bootstrapped with 100% organic growth (SEO + product-led growth)
- Available in 8 languages (Portuguese, Spanish, French, Russian, more)
- Trusted by thousands of schools and universities
We use multiple models depending on the task. Our main AI detection engine is based on a fine-tuned transformer model optimized for classification of synthetic vs human-written text.
- Model Base:
DeBERTa-v3-largeandRoBERTa-large(for earlier versions) - Task: Binary classification (
AIvsHuman) - Tokenization: Hugging Face Tokenizer (byte-level BPE)
- Output Method: Mean-pooled representation + linear classification head → probability score
- Inference: Optimized for short/medium-length texts (100–800 tokens)
- Balanced dataset of human-written and AI-generated text:
- Human: Wikipedia, Reddit ELI5, scientific abstracts, news articles
- AI: GPT-2, GPT-3.5, GPT-4, Claude, LLaMA, and other LLMs
- Custom datasets sourced and curated with real-world examples, including:
- Student assignments
- Blog posts
- Research summaries
- Multilingual support trained on non-English corpora (Portuguese, Spanish, French, etc.)
- F1 score: 0.92 (on balanced benchmark)
- False positives: <1.8% on academic test sets
- Latency: <600ms per 400 words (server-side inference)
- Plagiarism Checker: Compares input against web and internal databases for similarity.
- Fact Checker: Uses retrieval-based QA and semantic matching to flag false or outdated claims.
- API Access: High-volume usage for universities, publishers, and LMS platforms.
- Dashboard: Allows users to manage checks, share results, and access history.
- Frontend: Next.js
- Backend: Django (Python)
- Database: PostgreSQL + Firebase Firestore
- AI Serving: PyTorch + Hugging Face Transformers
- Cloud Infrastructure: Google Cloud Platform (GCP), Cloudflare CDN
- Languages Supported: English, Portuguese, Spanish, French, Russian, Uzbek (coming)
We are part of the following global startup ecosystems:
- Microsoft for Startups
- Google Cloud for Startups
- AWS Activate
- NVIDIA Inception
- Cloudflare Startup Program
These programs provide technical, cloud, and GTM support to help us scale globally.
This repository is intended for project overview and public information only.
The full source code and model weights are proprietary and are not shared in this public repo.
For partnership or demo requests, please contact:
📧 abdulla@detecting-ai.com