SecuriLayer: Using triple-layer AI (ONNX + Gemini/Llama) for real-time scam detection in community platforms

Gemini API Documentation
, ,
1

Hi everyone,

I’ve been building SecuriLayer — an AI-powered scam detection platform
for Telegram, Discord, and Slack communities.

The technical approach might be interesting here:
We use a triple-layer inference pipeline:

  1. ONNX + LightGBM for fast ML classification (<100ms)
  2. Llama 3.2 (via Ollama) for natural language threat explanations
  3. A context-aware assistant (Vigil) that understands the user’s specific risk profile

I’m considering integrating Gemini API as an additional explanation layer
for better multilingual support.

Would love technical feedback from anyone who has worked on similar
real-time threat detection pipelines or LLM explanation systems.

securilayer.dev — free tier available if anyone wants to test the detection quality.