About FaceIQ

A local-first AI face analysis tool. Age, gender, emotion — processed on your machine, powered by classical vision networks and a modern LLM.

How precision works

The Caffe model classifies faces into 8 broad buckets (0–2, 4–6…60+). Groq's vision LLM then receives a 224×224 crop and returns a tight 4–5 year range based on skin texture, bone structure, and facial fat distribution. The two results are merged to give you precise estimates like "22–26".

Tech stack

FastAPIPython async HTTP server with OpenAPI docs

OpenCV SSDResNet-10 single-shot face detector

Caffe (Levi-Hassner)8-bucket age & gender classification nets

Groq LLMllama-4-scout-17b vision model for ±3yr precision

Next.js 16React App Router frontend with Turbopack

WebSocketBinary streaming at 10fps with backpressure

Privacy

No images are stored anywhere. The frontend holds frames in memory only during active analysis. Groq receives a single 224×224 JPEG crop per face — no full image, no metadata, no identifiers. The backend holds a rolling in-memory history of results (not images) that resets on restart.