Case studies
No resume language, no buzzwords. Each project below is a real system I designed, built, and operate — problem, build, result.
JARVIS — Private Voice AI Assistant
Problem I wanted a voice AI assistant that could control my homelab, manage schedules, and respond hands-free — but every SaaS solution required monthly fees, sent data to third parties, and couldn't integrate with local infrastructure.
What I built A full-stack voice AI pipeline running entirely on self-hosted hardware — Whisper speech recognition on an RTX 3090, Gemma 12B via Ollama for local LLM reasoning, Kokoro TTS for natural voice output — all routed through a Telegram gateway with smart-home integration. The companion Flutter app is on TestFlight, wrapping the backend in a clean iOS interface with voice-first UX.
Result A 24/7 voice AI assistant with sub-second response times, zero monthly SaaS costs, full homelab integration, and a Flutter app ready for App Store distribution.
2-min before/after demo — coming soon
See the code →
Automated Futures Trading Pipeline
Problem Manual futures trading is emotionally exhausting, requires constant screen time, and reacts too slowly to market micro-moves. I needed a system that could execute strategies without me in front of the terminal.
What I built An automated MES Micro E-Mini S&P 500 trading pipeline pairing Sierra Chart's professional-grade charting and order routing with custom Python strategy scripts — all running on dedicated Proxmox VMs with redundant connectivity and risk-management guardrails.
Result A fully autonomous trading system that executes predefined strategies during market hours, managed remotely through secure VPN access, with zero manual intervention required.
2-min before/after demo — coming soon
See the code →
Homelab AI Superstack
Problem Cloud GPU instances are expensive, data leaves your network, and you can't iterate fast when every experiment costs per-minute compute. I needed local AI infrastructure that could handle speech-to-text, LLM inference, TTS, and image generation at production quality.
What I built A multi-node Proxmox cluster anchored by an RTX 3090-powered Ubuntu server running Ollama, OpenWebUI, Whisper, Kokoro TTS, ComfyUI, and SearXNG — all connected through VLAN-segmented networking with ZFS-based storage, Wazuh SIEM security monitoring, and a 4-tier backup strategy.
Result Production-grade local AI infrastructure supporting voice pipelines, image generation, code assistance, and web search — all running on self-built hardware with full data sovereignty.
2-min before/after demo — coming soon
See the code →