EmbeddingGemma 2 beats Mac Spotlight 30/30 to 5/30 in offline search

A developer benchmarked Google's free EmbeddingGemma 2 embedding model on a Mac: it found the right file first in 30 of 30 queries versus 5 for Spotlight, fully offline at about 34 ms per query. A face-search app built on it found 114 of 121 family appearances with zero wrong names.

• EmbeddingGemma 2: 30/30 correct files vs Spotlight's 5/30, offline ~34 ms
• 62 files indexed in 5 seconds, no vector database required
• Face search: 114/121 appearances found, 0 wrong names after 5 clicks
• Gemma 4 E4B answers from files 11/11 in ~2 s, 6.5 GB peak memory