How AIRadars Works
Understand how data flows through our on-premise AI security platform. All processing happens within your infrastructure — no external API calls.
Platform Data Flow
AIRadars ingests data from multiple sources — documents, logs, and AI agent events — processes them through specialized engines, and outputs actionable intelligence to your dashboard.
Documents
PDF, DOCX, EML files uploaded for classification
Log Sources
Syslog, Windows Events, and API logs for behavior analysis
Agent Events
AI agent prompts and tool calls for governance
Document Classification Pipeline
Our classification engine uses a multi-layer approach combining transformer-based ML models with pattern matching for maximum detection accuracy. The entire pipeline runs on-premise with latency under 100ms per document.
Document Upload
Files are uploaded via API or web interface. Supported formats include PDF, DOCX, EML, MSG, and plain text.
Text Extraction
Content is extracted using pypdf, python-docx, or OCR for scanned documents.
Preprocessing
Text is normalized, tokenized, and prepared for ML inference.
Dual Detection
ML models (DistilBERT/RoBERTa) and pattern matching (regex, checksums) run in parallel for maximum accuracy.
Result Merging
Detections from both methods are deduplicated and combined with confidence scores.
Risk Scoring
A weighted risk score (0-100) is calculated based on detected content categories.