# Positioning — Sellable "Hacker" Skills

## One-line pitch (use everywhere)

> **I get local AI running on your own hardware — GPU passthrough, multi-host Ollama routing, and self-hosted LLMs that never leave your network.**

## Longer bio (reusable paragraph)

I build and hack homelab infrastructure — self-hosted AI/LLM deployments, multi-GPU routing, and embedded/hardware hacking. I've taken a retired Dell R630 server from stock to a 24/7 Ollama inference box (Quadro M4000 with a CUDA→Vulkan migration after Ollama dropped Maxwell support), and I run a 4-lane GPU routing architecture across three machines — an RTX 4080 SUPER, an RTX 3070, and that R630 — with a single model standard, per-lane latency/VRAM routing, and 100K-token context on 16GB of VRAM with zero CPU offload. On the hardware side I've rooted and tuned Android TV boxes (Amlogic A/B, Magisk, NetherSX2 emulation) and built embedded voice-assistant firmware on ESP32-S3. If you want AI running on hardware you own — instead of renting a cloud API — I'm the person who makes it work.

## What problems I solve (fastest → slowest to a paid gig)

| Skill | Problem solved | Who pays | Speed to gig |
|---|---|---|---|
| **Self-hosted AI / GPU passthrough** | "I want to run a local LLM but GPU/Ollama won't work" | startups, homelabbers, privacy-conscious devs | 🔥 fastest |
| **Homelab / sysadmin** | Proxmox, Docker, ZFS, backups, networking | small business, homelabbers | fast |
| **Hardware / embedded hacking** | Android TV rooting, ESP32 firmware, ADB automation | hobbyists, product makers | medium |
| **Defensive security / vuln assessment** | authorized pentest, app security review | SMBs, SaaS founders | slower (needs scope/authorization) |

## Positioning decision

**Lead with self-hosted AI + GPU passthrough.** It's the highest-demand, lowest-friction,
legally-clean lane, and it's fully documented in my own infrastructure. The "hacker" credibility
comes through in the hardware/embedded work (SK4 Pro rooting, ESP32 firmware) and the defensive
security stack — those are the *proof*, not the storefront.

## Why now

Self-hosting AI is exploding (privacy, cost, data control) but GPU/Ollama passthrough is exactly
where most people get stuck. That's a gap I've already solved three times on my own hardware.