My personal site. Static, fast, and boring where it matters.

Live: salman-ch.netlify.app

$ whoami
Software engineer · 9D Technologies · Earth

---

Why I rewrote it

The old version was a Gatsby 3 template from 2021. It still built, barely, but it no longer described what I do. After GSoC, Jenkins releases and a few projects I actually care about, the site was two years behind.

So I threw it out and started from an empty folder. The rules I set:

1. No framework runtime in the browser. Plain HTML, a little CSS, small scripts that only wake up when their section is on screen. The home page ships ~36 KB of HTML, ~13 KB of CSS and ~5 KB of external JS after gzip, plus ~75 KB of self-hosted fonts.
2. Nothing typed by hand that a machine can check. PR counts and statuses come from the GitHub API at build time.
3. Content is data. Every word on the page lives in src/data/, split by topic. No CMS. See src/data/README.md.
4. Works without JavaScript. Scripts only add behaviour on top. Turn JS off and every section still renders.

What's on the page

Thing	How it works
Hero terminal	A small shell. help, ask <question>, neofetch, git log, now, agent, open x, sudo hire salman. Tab completion and history.
LLM agent replay	Replays the control flow of signoz-ai-sre: rules for known failures, a tool-calling LLM for unknown ones, one safety check for both. Four scenarios including a prompt-injection what-if; switch Gemini / Claude / GPT. Data in src/data/ai.ts.
Ask my résumé	BM25 over /rag.json, which is generated from src/data/ on every build. Runs in the browser, loads on first question, answers only with cited lines. No model call, nothing to hallucinate.
Tail-sampling simulator	Simulated CI traffic through the PoS-OTel policies. Toggles rewrite the collector YAML live.
PDF drop test	Drop a file: SHA-256, page count and PDF version computed in the tab with Web Crypto. Proves the zero-upload claim.
Changelog	git log --graph rendering of src/data/changelog.ts, filterable, with stable FNV-1a short hashes.
Skills, with proof	Click a skill, get the links that prove it.
Experience as a trace	Each role is a span in a Jaeger-style waterfall. Bars are positioned from start and end months, so it redraws itself as time passes. Rows are <details>, no JS needed.
Open source	Every PR row links out. Statuses are synced from GitHub on each build, then filterable (merged, in review, reported).
Ezvor demo	Sliders feed a weighted sum and an FNV-1a hash of the inputs. Same input, same score, same hash. That's the whole point of Ezvor's readiness engine.
⌘K palette	Keyboard menu for navigation, links and copying my email. / opens it too.
For machines	/llms.txt, /rss.xml, sitemap, and JSON-LD Person schema.

Stack

• Astro 7, static output
• TypeScript, strict
• Hand-written CSS with custom properties for light and dark themes. No Tailwind.
• Fonts: Geist, Geist Mono and Kalam, self-hosted via Fontsource, latin subsets preloaded
• Netlify for hosting, with security headers and redirects in netlify.toml

Run it

Needs Node 22.12 or newer.

npm install
npm run dev       # http://localhost:4321
npm run build     # static output in dist/
npm run preview   # serve dist/ locally

In VS Code, press F5 and pick "Dev server".

Where things live

src/
├── data/
│   ├── site.ts         profile, links, now.json, credentials, metrics
│   ├── experience.ts   roles → the career trace
│   ├── open-source.ts  PRs and issues by org
│   ├── projects.ts     Ezvor, PDF Scanner, project cards
│   ├── ai.ts           AI principles + agent scenarios
│   ├── skills.ts       skills with proof links
│   ├── changelog.ts    dated progress log (edit this most)
│   ├── awards.ts       awards and scholarships
│   ├── github.ts       build-time PR sync, cached, with fallback
│   └── README.md       the update guide
├── components/         one file per section (Hero, Trace, OpenSource, ...)
├── content/writing/    blog posts in Markdown (external posts link out)
├── layouts/Base.astro  head, SEO, theme bootstrap
├── scripts/retrieve.ts BM25 retriever, loaded on demand
├── pages/              index, writing, 404, rss.xml, llms.txt, rag.json
└── styles/global.css   tokens, reset, shared primitives
public/                 resume.pdf, og.png, favicon, robots.txt
scripts/og.mjs          regenerates the 1200x630 social card

Changing things

• Anything on the page: edit the matching file in src/data/. The guide in src/data/README.md has a table of which file to touch.
• A new post: add a Markdown file to src/content/writing/. For a post hosted elsewhere, set external and publisher in the frontmatter and it links out.
• Résumé: replace public/resume.pdf. /resume redirects to it.
• Social card: edit scripts/og.mjs, then run node scripts/og.mjs.

The GitHub sync

src/data/github.ts runs once per build. It searches every PR I've opened outside my own repos, keeps the ones in upstream orgs, and works out:

• the merged count and org count shown on the page
• the live status of each PR listed in profile.ts

If the API is rate limited or down, the build doesn't fail. It falls back to the statuses written in profile.ts and shows "status from last sync".

Two optional Netlify settings:

• GITHUB_TOKEN as an environment variable, for a higher rate limit
• a daily build hook, so a PR that merges overnight shows as merged the next morning

Deploy

Netlify builds on push. The settings are in netlify.toml: build with npm run build, publish dist.

License

MIT. Take the code if it helps. The words and the résumé are mine.