JCJim Chundevalelinaccidentalcomplexity.hashnode.dev·Jul 5 · 19 min readPart 3: Open WebUI, a One-Sided Load Balancer, and a Bug Hiding One Field DeepTwo Birds, One Qdrant With the core database built and my own homelab knowledge ingested, I had a second idea. Open WebUI, the chat front end I already run for my Hailo and Gemma models, has its own b00
JCJim Chundevalelinaccidentalcomplexity.hashnode.dev·Jul 4 · 8 min readPart 2: Herding Files Across Five Pis, Dodging Junk, & Teaching Qdrant What MattersThe Plan Was Simple Qdrant was running. I had a collection. All I needed to do was feed it something. My homelab "knowledge" was scattered across five different Pis: config files here, Modelfiles ther00
JCJim Chundevalelinaccidentalcomplexity.hashnode.dev·Jul 3 · 9 min readPart 1: RAID, Rust, and a Very Stubborn Memory AllocatorThe Idea I’ve had a local AI stack running for a while now: Hermes on WhatsApp, Gemma 4 on CPU, two Hailo AI HAT+ boards handling NPU inference, and Open WebUI tying most of it together. It was good a00
JCJim Chundevalelinaccidentalcomplexity.hashnode.dev·Jun 20 · 26 min readPart 4: An Edge AI Circus - Running Gemma 4 on a Pi 5 with OllamaIntroduction This is my build log. The goal was simple: run Google's Gemma 4 models locally on a 16 GB Raspberry Pi 5 with Ollama. No cloud. No GPU. Just a small board on a desk doing real inference f00
JCJim Chundevalelinaccidentalcomplexity.hashnode.dev·Jun 7 · 17 min readPart 3: The Brain, Hermes Agent, WhatsApp, and the Tiny Model Reality Check Introduction Parts 1 and 2 built the infrastructure: two HAT Pis running hailo-ollama on Hailo-10H silicon, Open WebUI sitting in front of them behind a clean API, SearXNG handling private web search,00
