Aiscern — AI Content Detection Platform

Aiscern is an AI-generated content detection SaaS that identifies AI-produced text, images, audio, and video using a multi-model ensemble — combining HuggingFace transformers, Gemini 2.0 Flash, NVIDIA NIM, and a retrieval-augmented generation (RAG) pipeline for high-accuracy, explainable results.

🌐 Live: aiscern.com

---

Features

Modality	Models Used	Avg. Accuracy
Text	RoBERTa-base-openai-detector, Binoculars perplexity, Gemini	~95%
Image	ViT-based classifier, CLIP embeddings, pixel-integrity signals	~91%
Audio	wav2vec2 fine-tuned, spectral analysis, SynthID local	~88%
Video	NVIDIA NIM deepfake detection, frame-level ensemble	~85%

• RAG-augmented ensemble — retrieves similar patterns from pgvector for context-aware confidence
• Web scanner — full-page crawl with JavaScript rendering via Playwright
• Ephemeral scan mode — zero-history option for sensitive documents
• Batch detection — upload and process multiple files
• API access (Pro/Enterprise) — REST API with per-key rate limiting
• Admin dashboard — real-time pipeline stats, D1 queue monitoring, user management

---

Tech Stack

Layer	Technology
Frontend	Next.js 15 (App Router), TypeScript, Tailwind CSS
Auth	Clerk (JWT, OAuth)
Database	Supabase (PostgreSQL + pgvector)
Storage	Cloudflare R2
Edge DB	Cloudflare D1
Inference	HuggingFace Inference API, Gemini 2.0 Flash, NVIDIA NIM
Background jobs	Inngest
Rate limiting	Upstash Redis
Payments	XPay (PKR + international)
Logging	Pino (structured JSON)
CI/CD	GitHub Actions → Vercel

---

Quick Start

Prerequisites

• Node.js 20+
• npm 10+
• A Supabase project
• A Clerk application
• A Cloudflare account (R2 bucket + D1 database)

Installation

git clone https://github.com/anasali89ji/AI-SCERN.git
cd AI-SCERN/frontend
npm install --legacy-peer-deps

Environment Variables

Copy .env.example to .env.local and fill in all values:

cp .env.example .env.local

See Environment Variables section for full reference.

Database Setup

Apply Supabase migrations in order:

# In Supabase SQL Editor, run each file in /supabase/migrations/ in order:
# 1. v6_pipeline_schema.sql
# 2. hardening.sql
# 3. calibration_tables.sql
# 4. calibration_stats.sql
# 5. performance_indexes.sql
# 6. r2_storage_column.sql
# 7. feature_flags.sql
# 8. v8_api_key_sha256_migration.sql
# 9. v9_scraper_sessions.sql
# 10. forensic_pipeline.sql

Development

cd frontend
npm run dev

Open http://localhost:3000.

Production Build

cd frontend
npm run build
npm start

---

Environment Variables

Variable	Required	Description
NEXT_PUBLIC_SUPABASE_URL	✅	Supabase project URL
NEXT_PUBLIC_SUPABASE_ANON_KEY	✅	Supabase anon/public key
SUPABASE_SERVICE_ROLE_KEY	✅	Supabase service role — server only, never expose
NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY	✅	Clerk publishable key
CLERK_SECRET_KEY	✅	Clerk secret key — server only
GEMINI_API_KEY	✅	Google Gemini API key
GROK_API_KEY	optional	xAI Grok Vision (image detection primary)
OPENROUTER_API_KEY	optional	OpenRouter fallback
HUGGINGFACE_API_TOKEN	optional	HuggingFace Inference API
NVIDIA_API_KEY	optional	NVIDIA NIM (video deepfake)
R2_ACCOUNT_ID	✅	Cloudflare account ID
R2_ACCESS_KEY_ID	✅	R2 API key
R2_SECRET_ACCESS_KEY	✅	R2 API secret
R2_BUCKET_NAME	✅	R2 bucket name
CLOUDFLARE_D1_DATABASE_ID	✅	D1 database ID
UPSTASH_REDIS_REST_URL	✅ (prod)	Upstash Redis URL
UPSTASH_REDIS_REST_TOKEN	✅ (prod)	Upstash Redis token
INNGEST_EVENT_KEY	✅	Inngest event key
INNGEST_SIGNING_KEY	✅ (prod)	Inngest signing key
SENTRY_DSN	optional	Error tracking
INTERNAL_API_SECRET	✅	Server-to-server auth secret

Full reference: .env.example

---

Project Structure

AI-SCERN/
├── frontend/                  # Next.js 15 application
│   ├── app/
│   │   ├── (auth)/            # Login, signup pages
│   │   ├── (dashboard)/       # Detect, history, settings
│   │   ├── (marketing)/       # Landing, pricing, blog
│   │   ├── (legal)/           # Privacy, terms, contact
│   │   └── api/               # API routes
│   ├── components/            # Shared UI components
│   ├── lib/
│   │   ├── auth/              # Auth utilities
│   │   ├── inference/         # HF, Gemini, NVIDIA inference
│   │   ├── rag/               # RAG pipeline (pgvector)
│   │   ├── security/          # File validation, CSRF
│   │   ├── storage/           # Cloudflare R2
│   │   └── supabase/          # DB client
│   └── next.config.ts
├── admin/                     # Separate admin Next.js app
├── signal-worker/             # Python image analysis worker
├── supabase/migrations/       # Database migrations
├── .github/workflows/         # CI/CD pipelines
└── cf-pipeline/               # Cloudflare Worker pipeline

---

Benchmarks

Accuracy metrics are published on the Methodology page with per-modality AUC-ROC, precision, recall, and false-positive rates.

Key datasets used for evaluation:

• Text: PAN25, PERSUADE Corpus 2.0, M4 Benchmark
• Image: FakeFace, CIFAKE, GenImage
• Audio: ASVspoof 2019/2021, ADD 2023
• Video: FaceForensics++, DFDC Preview

---

API Reference

Full API documentation: aiscern.com/docs/api

# Example: Detect text via API
curl -X POST https://aiscern.com/api/v1/detect/text \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "Your content here"}'

---

Contributing

See CONTRIBUTING.md for guidelines.

---

Security

To report vulnerabilities, see SECURITY.md or email security@aiscern.com.

---

License

MIT — see LICENSE.

---

Citation

If you use Aiscern in research, please cite:

@software{aiscern2024,
  title   = {Aiscern: Multi-Modal AI Content Detection},
  author  = {Aiscern Team},
  year    = {2024},
  url     = {https://github.com/anasali89ji/AI-SCERN}
}