Snapshot: full project state

portfolio/01-r630-gpu-ollama.md

@@ -0,0 +1,35 @@
+# Portfolio — R630 GPU Passthrough → Always-On Ollama Inference
+
+## Problem
+
+A retired Dell R630 PowerEdge (service tag BXK1MR2) needed to become a 24/7 local
+LLM inference box for a fleet of self-hosted apps — instead of each app spinning up
+its own Ollama instance or renting cloud GPU time.
+
+## Approach
+
+- **Hardware**: R630 with a Quadro M4000 (8GB). Expanded RAM from 64GB → 117GB, and
+  carved out NVMe into a 32GB SLOG + 174GB L2ARC + 32.5GB swap for ZFS.
+- **GPU passthrough**: Installed Docker + `nvidia-toolkit`, attached the M4000 with
+  `--gpus all`, and stood up Ollama as a systemd service (`0.0.0.0:11434`, LAN-open,
+  `keep_alive 30m`, `max_loaded_models 1`).
+- **The gotcha nobody documents**: Ollama 0.34 *dropped CUDA for Maxwell* (compute 5.2
+  needs driver 570+, box has 550). CUDA silently fails → Ollama falls back to **Vulkan**.
+  Diagnosed it, kept Vulkan, and benchmarked it properly instead of assuming it was dead.
+
+## Result
+
+- `ornith-1.5:9b` (9B, Q4_K_M, 6.6GB) runs **100% GPU at 13.2 tok/s** with a 4K context —
+  comparable to an RTX 3070 at 64K context, on a $100 retired GPU.
+- The box is now the **always-on consolidation lane**: every server app (Tarro, PHOTON,
+  Signal Miner, Research Engine, any Flask/Node app with a default LLM) points at one
+  Ollama instead of each running its own.
+
+## Tools used
+
+Proxmox, Docker, nvidia-toolkit, Ollama, Vulkan, ZFS (SLOG/L2ARC), systemd, Dell iDRAC8.
+
+## Why it's sellable
+
+This is the exact problem people pay to solve: "I have a GPU, why won't Ollama use it?"
+The Maxwell/CUDA→Vulkan migration is a real, non-obvious fix most people burn hours on.

portfolio/02-multihost-gpu-routing.md

@@ -0,0 +1,40 @@
+# Portfolio — 4-Lane Multi-GPU LLM Routing Architecture
+
+## Problem
+
+Three GPU machines (an RTX 4080 SUPER, an RTX 3070, and a Quadro M4000) plus a MacBook
+were being used ad hoc, with every app guessing where to send its LLM calls. Result:
+GPU saturation, cold-model timeouts, and latency-critical requests queued behind batch work.
+
+## Approach
+
+Designed a **4-lane routing architecture** with a single model standard (`ornith-1.5:9b`)
+and an explicit per-lane map:
+
+| Lane | Host | GPU | Job |
+|---|---|---|---|
+| **premium** | RTX 4080 SUPER (16GB) | CUDA | mission-critical speed only |
+| **batch** | RTX 3070 (8GB, WiFi) | CUDA | stateful, long-context, vision/OCR/embeddings |
+| **permanent** | Quadro M4000 (8GB) | Vulkan | always-on consolidation for server apps |
+| **voice** | MacBook (unified) | Metal/MLX | local voice (gemma3:4b) |
+
+## Result
+
+- **100K-token context on 16GB VRAM with zero CPU offload** — baked `num_ctx 102400`,
+  `num_gpu 99`, `KV_CACHE_TYPE q4_0`, and `NUM_BATCH 2048` to push prefill from 308 →
+  **1867 tok/s** and decode to ~41 tok/s on the 4080.
+- Single shared GPU for Ollama + ComfyUI (image gen) via HyperSwap (program swapper),
+  so text inference and diffusion don't fight over VRAM.
+- Every consumer (TITAN, Astraea, WorkBrain, PHOTON, Honcho, 90+ cron jobs) mapped to the
+  lane that fits its latency/VRAM class.
+
+## Tools used
+
+Ollama (multi-host), CUDA + Vulkan, Flash Attention, KV-cache quantization, HyperSwap,
+ComfyUI, systemd, LAN routing.
+
+## Why it's sellable
+
+"Run multiple LLMs across multiple GPUs without paying a cloud provider" is a real,
+growing ask. The 100K-context-on-16GB result is a concrete, quantifiable win most
+consultants can't show.

portfolio/03-sk4pro-android-tv.md

@@ -0,0 +1,33 @@
+# Portfolio — Android TV Box Rooting + Emulation (SK4 Pro)
+
+## Problem
+
+Turn a UGOOS SK4 Pro Android TV box (Amlogic, Android 14) into a retro-emulation
+console — including PS2 emulation via NetherSX2 — which required root, BIOS/ROM
+staging, and core configuration.
+
+## Approach
+
+- **Device**: SK4 Pro, Amlogic SoC, Android 14 with A/B partitions.
+- **Root**: Pushed Magisk 30.7, generated a `magisk_patched` boot image, and staged the
+  patched `init_boot` partition for the A/B slot.
+- **Emulation**: Configured RetroArch PlayStation cores (needs BIOS + ROMs), staged
+  `scph5501.bin`-class PS1 BIOS and `.bin/.cue`/`.chd` ROMs, and prepped NetherSX2 (PS2)
+  with its required PS2 BIOS set (SCPH-39001/70012/77001/90001).
+- **Access**: ADB over network (port 5555) + SMB Samba share into `/sdcard` for ROM
+  management when the ADB device was still in "unauthorized" state.
+
+## Result
+
+A rooted, emulation-ready TV box with N64/PS1/PS2 paths staged, Jellyfin side-loaded for
+media, and a clean ROM/BIOS layout — all managed remotely over the LAN.
+
+## Tools used
+
+Amlogic A/B flashing, Magisk, ADB, RetroArch, NetherSX2, SMB/Samba, Jellyfin.
+
+## Why it's sellable
+
+This is the "hacker" proof — real firmware/boot-image work on Android hardware, not just
+app installs. Same skillset applies to ESP32-S3 embedded firmware (I also built a voice
+assistant on ESP32-S3). Demonstrates I can go below the OS layer when the job needs it.