JEWRADAR

Multimodal Frontier Embeddings

Gemini Embedding 2 encodes your face into a 3,072-dimensional latent space — frontier-model power with zero guardrails

1,691-Face Reference Corpus

Pre-computed embedding matrix loaded in-memory. Top-30 nearest neighbor retrieval via brute-force cosine similarity in <50ms

Calibrated Sigmoid Scoring

Raw similarity mapped through a steep sigmoid (midpoint 0.722, k=200) — tiny embedding shifts produce clean score separation

Guardrail-Free Architecture

No content policy. No wrapper. Raw vector math on embeddings you own. This is what AI looks like without corporate training wheels