# Contact Scout

> Contact extraction API with a self-hosted vision-model fallback

Status: active · 2026 · AI

**Problem.** Contact pages hide emails in images and JavaScript; classic scrapers return nothing.

**Approach.** A dual-engine pipeline: fast HTML extraction first, and when a page yields nothing, a self-hosted vision LLM reads the rendered screenshot. Emits E.164 phones, IČO/DIČ, socials; HMAC-signed webhooks; Prometheus metrics.

**Result.** In daily use; actively maintained.

**Stack.** FastAPI, Scrapling, Botasaurus, llama.cpp, Qwen3-VL-2B

**Links.** [Page](https://vladoportos.sk/u/contactscout)

A project by [Vladimír Strýček (VladoPortos)](https://vladoportos.sk/), Cloud & DevOps Engineer in Bratislava, Slovakia. Runs on the homelab: gpu. Full CV: https://vladoportos.sk/cv.md
