mirror of https://github.com/blackboxprogramming/alexa-amundson-resume.git synced 2026-03-18 05:34:08 -05:00

Files

Alexa Amundson ec7b1445b5 kpi: auto-update metrics 2026-03-13

RoadChain-SHA2048: c645c1292ab1555e
RoadChain-Identity: alexa@sovereign
RoadChain-Full: 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

2026-03-13 23:16:12 -05:00

2.3 KiB

Raw Blame History

Alexa Amundson

Systems Engineer

amundsonalexa@gmail.com | github.com/blackboxprogramming

Summary

When your production fleet is single-board computers, every kernel parameter matters. Tuned CPU governors, stabilized voltage, integrated PCIe AI accelerators, and squeezed production workloads from hardware that fits in your hand.

Experience

BlackRoad OS | Founder & Systems Engineer | 2025–Present

The Constraint: Maximum Work from Minimum Hardware

A Pi 5 has 8 GB RAM, a quad-core ARM, and a 30W power budget. It needs to run Docker, Ollama, Nginx, PostgreSQL, and 50+ systemd services simultaneously
Tuned swappiness to 10, dirty_ratio to 40, applied conservative CPU governors, capped frequency to 2 GHz — workloads stable, temperatures safe
GPU memory reduced from 256MB to 16MB on headless nodes — freed RAM for actual compute. Disabled cups, rpcbind, nfs, lightdm across fleet

The Integration: Making Hardware Talk

2x Hailo-8 NPU via PCIe — installed drivers, firmware, verified /dev/hailo0 on both nodes. 52 TOPS of AI acceleration, zero cloud cost
NVMe SSD on Octavia (1TB) — faster I/O for Gitea, Docker images, and model weights. USB peripherals: UART, keyboards, microphones, OLED displays
Overclock on one node caused undervoltage (0.75V) — removed overclock, tuned config.txt, recovered +95mV. Fleet-wide voltage monitoring deployed

The Discipline: 256 Services, Zero Chaos

256 systemd services and 35 timers across fleet — each one has a purpose, a health check, and an owner
Self-healing watchdogs restart failed services. Power monitoring logs every 5 minutes. Everything persistent across reboots via sysctl.d and tmpfiles.d

Technical Skills

Linux kernel, systemd, sysctl, PCIe, I2C, GPIO, Hailo-8, NVMe, Bash, Python

Metrics

Metric	Value	Source
Systemd Services	live	services.sh — systemctl list-units via SSH
Systemd Timers	live	services.sh — systemctl list-timers via SSH
Fleet Nodes	live	fleet.sh — SSH probe to all nodes
Avg Temp	live	fleet.sh — /sys/class/thermal via SSH
Fleet RAM (MB)	live	fleet.sh — /proc/meminfo via SSH
Fleet Storage (GB)	live	fleet.sh — df via SSH

2.3 KiB Raw Blame History Unescape Escape