Javi is a voice assistant and agent orchestrator running entirely on my own GPU. No cloud APIs. No data leaving the machine.
Under the hood
12B LLM on Ollama (ROCm) with native tool calling: it plans, executes tools, checks results and answers — unsupervised.
Whisper large-v3-turbo transcribes in 0.2s; XTTS-v2 speaks the reply in ~1s per sentence. Conversation feels live.
Self-hosted meta-search (SearXNG), deterministic deep-research and mandatory citations on every factual claim.
Automated job radar on n8n: 6 sources scanned, ranked by salary and seniority, summarized out loud every morning.
SSRF guards, CORS scoping, healthchecks, supply-chain pinning — and 67 tests keeping it honest.
Every container runs on my own hardware. Privacy isn't a feature flag — it's the architecture.
Architecture
Private repo — but I'll happily walk you through any of it on a call.
Built with
I'll walk you through the architecture, the trade-offs and the war stories.