---
title: "The Architecture of a Modern Personal Site"
date: "2025-02-06"
description: "A deep dive into the technical stacks, design philosophies, and implementation details of building a minimalist digital garden in the age of AI."
tags: ["Architecture", "Next.js", "Design", "System"]
---
# Introduction
In an era dominated by algorithmic feeds and ephemeral content, the personal website remains the last bastion of digital sovereignty. It is a space where the signal-to-noise ratio is controlled entirely by the author.
This post serves as **stress test** for the typography and layout of this blog engine. We need to verify that long-form content is readable, that the table of contents tracks correctly, and that the marginalia (side notes) function as intended.
## The Tech Stack
We opted for a "bleeding-edge but stable" approach. The core constraints were:
1. **Performance**: Sub-100ms navigation.
2. **Simplicity**: No heavy client-side hydration unless necessary.
3. **Ownership**: Markdown-backed content.
Markdown ensures that the content is portable. If this site moves to a completely different stack in 5 years, the content remains valid.
### Next.js & React Server Components
We utilize React Server Components (RSC) to render the MDX content entirely on the server. This means zero JavaScript is shipped for the blog post content itself.
```typescript
// app/blog/[slug]/page.tsx
export default async function BlogPost({ params }: Props) {
const { slug } = await params;
const post = getPostBySlug(slug);
return (
);
}
```
## Typography & Rhythm
Good typography is invisible. It guides the eye without drawing attention to itself. We use **Inter** for its tall x-height and readability on screens.
The vertical rhythm is maintained using `prose` (Tailwind Typography), ensuring that margins between paragraphs, headings, and lists are consistent.
### List Styles
* **Unordered lists** are used for collections of related items.
* **Ordered lists** imply a sequence or ranking.
* **Nested lists** should handle indentation gracefully.
1. First item
2. Second item
* Sub-item A
* Sub-item B
3. Third item
## Handling Complexity
What happens when we introduce complex diagrams or citations?
The system uses a custom citation component that links to a bibliography at the end. This is inspired by Tufte's design principles.
> "Clutter and confusion are failures of design, not attributes of information." — Edward Tufte
## Deep Dive: The Graph Network
Let's discuss the graph neural network work mentioned in my resume. When dealing with multi-agent systems, the topology of interaction is often unknown.
### Inference Challenge
Inferring the graph $G = (V, E)$ from a set of trajectories $T$ is an NP-hard problem in its general form. However, by restricting the hypothesis space to local interactions, we can approximate the adjacency matrix.
The attention mechanism acts as a "soft" adjacency matrix, allowing the model to learn weights between agents dynamically.
We developed a novel **Fast Graph Attention** layer that allows for:
1. **Batching**: Processing heavily disjoint graphs in parallel.
2. **Sparsity**: Pruning edges with low attention scores during training.
```python
class FastGAT(nn.Module):
def forward(self, x, adj):
# Efficient sparse matrix multiplication
return spmm(adj, self.W(x))
```
## Conclusion
This layout seems to handle various content types well. The sticky TOC on the left should have tracked your progress through *The Tech Stack*, *Typography*, and *Deep Dive* sections.
The marginalia should have appeared to the right of the relevant paragraphs without overlapping the main text.
[1] Edward Tufte, *Envisioning Information*, Graphics Press, 1990.