HomeBlogDemos

“The details are not the details. They make the design.”— Eames

abhishekghoshsde@gmail.com
Join the DispatchNo spam · Unsub anytime
HomeBlogDemosGitHub
© 2026 CodeNuclei. All rights reserved.
Writing/Engineering
Engineering

Demos next to the blog

Every side project ends up with a demo that needs to live somewhere. This site hosts them under /demos/<name> on the same domain as the writing.

CN
Codenuclei@codenuclei
November 11, 2026·1 min read

Every side project ends up with the same problem: a demo that needs to be somewhere, and a blog post that needs to link to it. The usual answer is a scattering of hostnames — three Vercel projects, a Netlify drop, a screenshot on imgur because the demo's cold start was too slow to share.

This site does it the boring way: demos live under /demos/<name> on the same domain as the writing.

How it works

Each demo is a Next.js route. The registry in src/lib/demos.ts lists what exists; the /demos page renders cards from it. Adding a demo is:

  1. Create src/app/demos/<slug>/page.tsx (and a client component if it needs state).
  2. Add an entry to src/lib/demos.ts.
  3. Commit. The card appears on /demos and the route is statically generated at build time.

No separate hosting, no separate deploy, no separate DNS. The demo and the post that explains it share a cache, a domain and a build.

What this buys

  • One build. bun run build produces the blog and every demo in a single static output.
  • One link. A post can deep-link to /demos/theme-explorer and the reader never leaves the site.
  • One theme. Demos inherit the same matte-white tokens as the blog, so a demo embedded in a post looks like part of the post.

The theme explorer walks the palette; the ui kit shows the stock shadcn components on it.

CN

Codenuclei

Author

Real-time graphics, GPU inference optimization, and bespoke mathematical UI.

All Articles
Continue Reading
Previous Article

Continuous Batching, PagedAttention, and Speculative Decoding in High-Throughput LLM Serving

How virtual memory paging eliminates KV-cache fragmentation, while draft-model speculative decoding breaks the sequential token latency wall on Modern GPUs.

6 min read
Next Article

Dissecting Memory Bandwidth & SRAM Tiling in Triton: From GEMM to FlashAttention-3

Why LLM inference is strictly memory-bound in decoding, how HBM3e vs SRAM bandwidth dictates latency, and writing a fused Triton kernel with online softmax.

7 min read
Back to all publications
Contents2 sections
How it worksWhat this buys
Geist Sans