Everything I've built and run — filter by domain or search by tech. Each card opens a case study: problem → architecture → decisions → measured outcome.
A working replication of NVIDIA's “Small Language Models are the Future of Agentic AI” — their six-step LLM→SLM conversion pipeline, run against my own agent telemetry. Replaying every routing decision in hindsight put 57% of local-tier spend within reach of a 35B model on my own hardware, inside the paper's predicted 40–70% band.
A single-node k3s cluster running Grafana, Prometheus, Loki and ArgoCD — GitOps-deployed, 95.2% uptime, with a Discord incident-analysis bot. Powers the live metrics on this site.
A from-scratch SwiftUI music player — 10-band parametric EQ, 364 tests, zero third-party dependencies — backed by a Python/yt-dlp streaming service.
Measurement-based audiophile write-ups — gear testing, target curves, and the parametric-EQ profiles that came out of building Aria.
Service-mesh observability on k3s — metrics, logs, traces, and AI incident analysis — for $5.12/mo, 60% under serverless.
GitOps observability: a 20-panel Grafana dashboard and 7 alert rules across 5 services, every piece declared in Git and deployed by ArgoCD.
A PaperMC server wired into the cluster's observability stack: 95.2% uptime, TPS-drop alerts, and a 10-command Discord bot.
Serverless health checks every 15 min with 5-minute dedup windows that stop Discord alert storms.
This site: a vanilla-JS PWA with no runtime dependencies — Lighthouse 95+, WCAG 2.1 accessible, hosted free on Cloudflare Pages.