Projects.

Things I have shipped. Click through to use them, read how they were built, or grab the source.

3 projects · 2 live · ← → to browse, ⏎ to open

2026

The bench

Local LLM inference measured on one RTX 5090: speculative decoding, FP8 KV cache, concurrency, with the run logs.

350tok/s
code edit, n-gram spec: 4.65×
vs baseline: 656tok/s
16-way aggregate

AI
Benchmarks
vLLM
Qwen3-32B-AWQ
Python

Visit → Details →

How this site is built

This site: static Astro, build-time SVG charts, a sky that keeps real time, and a few KB of JS.

~6KB
client JS: 0
trackers: 0
frameworks shipped

Web
Design
Astro
TypeScript
Cloudflare Pages

Visit joeguglielmucci.com ↗ Details →

Mimiic

Plain-English patent search: AI reads the claims and a novelty report shows what's taken and what's open.

380,934
patents, one search: 0.56s
search time: $29
per report

AI
Web
Next.js
USPTO full text
Stripe

Visit mimiic.app ↗ Details →