Full-Stack Developer · Local AI Engineer Morelos, México · Open to remote roles · Spanish (native) / English (advanced)
I build web apps (React, Node.js, PostgreSQL, AWS) and desktop apps (Tauri, Rust) that run AI models on the user's own hardware: no cloud, no subscription, private by default.
- Python → Rust migration (LiveAudio): moved audio capture, voice activity detection, WebSocket broadcasting and process supervision to a native Rust workspace. Idle memory went from ~959 MB to ~73 MB and cold start from ~2.8 s to ~17 ms. See benchmarks →
- Real-time local speech recognition: Faster-Whisper
smallrunning at ~25× real-time on an RTX 3060, streamed as live subtitles into OBS. - Production full-stack work (RaukeIT, 2025): image-sharing platform and product catalog with Stripe payments, JWT auth, AWS S3 uploads, PostgreSQL and MongoDB.
| Project | Stack | What it does |
|---|---|---|
| LiveAudio | Rust Tauri 2 Faster-Whisper ONNX WebSocket |
Real-time local subtitles for streamers. Rust core + supervised Python ASR worker, OBS overlay, Windows/Linux. |
| OpenCohost | Tauri 2 Rust React Python Ollama |
Local AI co-host for streamers: sees the screen (vision model), listens (STT), talks back (TTS) and remembers conversations. Viewer chat never leaves the machine. |
| Brick.draw | Tauri React TypeScript Rust |
Lightweight layered drawing app with optional local Stable Diffusion XL enhancement. |
| EngramView | Tauri React TypeScript |
Read-only desktop viewer for local Engram memory stores. |
| Floppy (private) | Python CUDA DBSCAN SQLite |
GPU computer-vision pipeline for unsupervised face clustering at scale, with explicit VRAM management. |
Web : React, TypeScript, Next.js, Svelte, Tailwind
Backend : Node.js, Express, Python, PostgreSQL, MongoDB, SQLite, Stripe, JWT
Desktop : Tauri 2, Rust (Tokio, CPAL, ONNX Runtime)
Local AI : Faster-Whisper, Silero VAD, Ollama, Qwen-VL, Qwen Image 2.1, PiperTTS, convaiinnovations/laya, CUDA
Infra : AWS (EC2, S3), Cloudflare, Vercel, Linux, Git