bourbask.github.io

Personal portfolio and tech watch — built with Rust, Leptos, and WebAssembly.

Live: www.bourbasquetkev.in — also served at bourbask.github.io

---

What's in here

• Portfolio (about, skills, projects, contact)
• Blog with article routing
• Tech Watch (/veille) — daily news feed from RSS/HN, a weekly editorial synthesis, and a dedicated AI watch with its own brief; auto-updated via GitHub Actions
• Bilingual (FR/EN) with persistent language preference
• Dark/light theme with system preference detection, no flash on load
• Printable CV generator

---

Documentation

In-depth docs live in docs/:

Doc	Covers
veille-pipeline.md	Tech watch pipeline — sources, scoring, AI track, synthesis, schema
testing.md	Token-free pipeline tests (faked client, network guard)
architecture.md	Frontend Rust/Leptos/WASM — components, services, routing
deployment.md	CI/CD, GitHub Pages, custom domain
design-system.md	Design tokens, typography, accessibility
dev-setup.md	Local setup, commands, troubleshooting

---

Stack

Layer	Tech
Language	Rust (stable)
Framework	Leptos 0.8 (CSR)
Bundler	Trunk 0.21.14
Target	wasm32-unknown-unknown
Styling	CSS3 with custom properties
News fetch + scoring	Python 3 + feedparser + Claude Haiku (daily, GitHub Actions)
News synthesis	Python 3 + Claude Sonnet (general: Monday; AI brief: Mon + Thu)
Pipeline tests	pytest — offline, no API key, zero AI tokens

---

Setup

Standard install (most systems)

# Install Rust via rustup
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
source ~/.cargo/env

# Add WASM target
rustup target add wasm32-unknown-unknown

# Install Trunk
cargo install --locked trunk

Arch / Manjaro

The rustup package from pacman installs shims in /usr/bin (cargo, rustc) that point to rustup itself. Running cargo without a configured toolchain causes infinite recursion.

# Check if this is your situation
file /usr/bin/cargo   # → "symbolic link to rustup" = you have this issue

# Fix: install the stable toolchain and set it as default
rustup toolchain install stable
rustup target add wasm32-unknown-unknown
rustup default stable

# Verify
cargo --version   # should print "cargo 1.x.x ..."

# Install Trunk — via AUR (faster, no compile)
yay -S trunk
# or
paru -S trunk

# Alternatively, compile from source (takes ~5-10 min)
cargo install --locked trunk

Do not use rustup toolchain link system /usr. It links the shim as the toolchain, causing the same recursion.

---

Development

# Clone
git clone https://github.com/bourbask/bourbask.github.io.git
cd bourbask.github.io

# Start dev server (hot reload at http://127.0.0.1:9999)
trunk serve

# Production build
trunk build --release
# Output: dist/

Useful commands

trunk serve            # dev server with hot reload (http://127.0.0.1:9999)
trunk build            # debug build
trunk build --release  # optimized build for deployment

cargo clippy --release --target wasm32-unknown-unknown  # linting (matches CI)
cargo fmt              # formatting
cargo check            # fast type check (no binary output)

---

Quality checks

The CI pipeline runs accessibility, performance, HTML validity, and security header checks automatically after each deploy to main. Run the same checks locally before pushing to catch issues early.

Install tools (once)

npm install -g @lhci/cli pa11y

Docker is optional — needed only for the W3C HTML validator against localhost (the CI run uses the W3C public API instead).

Run before pushing

./scripts/quality_check.sh

The script starts trunk serve if not already running, runs all checks against http://localhost:9999, then exits and cleans up.

# Partial runs
./scripts/quality_check.sh --skip-w3c          # no Docker required
./scripts/quality_check.sh --skip-lhci         # skip Lighthouse
./scripts/quality_check.sh --skip-a11y         # skip pa11y

What each check covers

Check	Tool	Fails on
Performance, SEO, best practices	Lighthouse CI (@lhci/cli)	perf < 0.80, SEO < 0.90
Accessibility WCAG 2.1 AA	pa11y + axe-core	any WCAG2AA error
HTML validity	W3C Nu Validator	any HTML error
HTTP security headers	curl	missing X-Frame-Options, X-Content-Type-Options, Strict-Transport-Security

Thresholds are defined in .lighthouserc.json.

CI equivalent

The same checks run in .github/workflows/quality.yml, triggered automatically after each successful deploy to main. Results are available in the Actions tab; Lighthouse reports are uploaded as artifacts (30-day retention).

---

Project structure

src/
├── lib.rs                    # WASM entry point
├── app.rs                    # Router and top-level contexts
├── components/
│   ├── hero.rs
│   ├── about.rs
│   ├── skills.rs
│   ├── projects.rs
│   ├── interests.rs
│   ├── contact.rs
│   ├── navigation.rs
│   ├── mobile_nav.rs
│   ├── footer.rs
│   ├── not_found.rs
│   ├── veille.rs             # Tech watch page + synthesis detail
│   ├── blog/
│   │   ├── blog_page.rs
│   │   └── article_page.rs
│   └── ui/
│       └── cv_download.rs
├── services/
│   ├── i18n.rs               # Language switching + localStorage persistence
│   ├── theme.rs              # Dark/light theme
│   ├── storage.rs            # localStorage wrapper
│   ├── blog.rs
│   └── cv.rs
└── data/
    ├── cv.rs
    ├── articles/
    └── translations/
        ├── en.rs
        └── fr.rs

scripts/
├── fetch_news.py             # RSS + HackerNews → public/news.json (no AI)
├── score_articles.py         # Per-domain scoring + selection (Claude Haiku)
├── synthesize_news.py        # Weekly general synthesis + dedicated AI brief (Claude Sonnet)
├── repair_synthesis_links.py # One-shot repair of source links in syntheses (no AI)
├── check_feeds.py            # Feed reachability smoke check (no AI)
├── discover_tools.py         # Open-source tool discovery (tools pipeline)
├── generate_article.py       # Blog article generation for discovered tools
├── tests/                    # pytest suite — offline, zero AI tokens
└── quality_check.sh          # Local quality checks (mirrors CI)

public/
├── news.json                 # Generated — do not edit manually
├── css/                      # Design system (variables, components, pages)
└── fonts/                    # Self-hosted Literata variable font (OFL-1.1)

.github/workflows/
├── deploy.yml                # Build + deploy to GH Pages on push to main
├── ci.yml                    # Clippy + cargo audit on develop / PR to main
├── news-pipeline.yml         # Daily fetch + score; general synthesis (Mon), AI brief (Mon+Thu)
├── tests.yml                 # pytest pipeline tests (no API key → zero tokens)
├── feeds-smoke.yml           # Weekly feed reachability check (no AI)
├── tools-pipeline.yml        # Weekly open-source tool discovery + article PR
└── quality.yml               # Post-deploy quality checks (Lighthouse, pa11y, W3C, headers)

---

Tech Watch (Veille)

The /veille page shows a daily news feed, a weekly editorial synthesis, and a dedicated AI watch (own tab + own brief). Everything is generated by news-pipeline.yml.

Daily — fetch + score (no synthesis):

1. fetch_news.py pulls HackerNews + ~45 RSS feeds across 6 domains (dev_stack, ai, security, health_science, business_market, architecture), classifies, and commits to public/news.json.
2. score_articles.py scores each domain with Claude Haiku and selects a general top 5 (AI excluded) plus a dedicated AI top.

Synthesis:

• General (Monday) — synthesize_news.py --track general writes a bilingual (FR + EN) editorial synthesis. AI-domain articles are excluded.
• AI brief (Monday + Thursday) — synthesize_news.py --track ai writes a short, token-lean, English-only brief (state of the art → implications → developer actions).
• Source links use srcref:<id> tokens swapped for exact feed URLs after generation (no hallucinated 404s). Hero illustrations pass a relevance gate.

Synthesis cards appear in the feed; clicking opens a full Markdown detail page. The "IA" tab isolates the AI watch.

Run locally (make):

make fetch        # RSS + HN, no AI
make score        # daily scoring (needs ANTHROPIC_API_KEY)
make synth        # general weekly synthesis
make synth-ai     # dedicated AI brief
make test         # pipeline tests — offline, no API key, zero tokens

Repair broken source links in already-published syntheses (deterministic, no AI):

python scripts/repair_synthesis_links.py --dry-run

See docs/veille-pipeline.md for the full pipeline and docs/testing.md for the token-free test design.

---

Deployment

Push to main triggers .github/workflows/deploy.yml, which builds with Trunk and deploys to GitHub Pages.

Custom domain: served at www.bourbasquetkev.in. A CNAME file (copied to the build root via data-trunk) keeps the domain pinned across Actions deploys. DNS and cohabitation with the homelab VPS are documented in docs/deployment.md.

The CI workflow (.github/workflows/ci.yml) runs on every push to develop and every PR to main:

• cargo clippy --release --target wasm32-unknown-unknown -D warnings
• cargo audit (RUSTSEC advisory check)

Important: Do not set data-wasm-opt in index.html. wasm-opt without --enable-reference-types corrupts the wasm-bindgen externref table at runtime (blank page).

---

i18n

Translations live in src/data/translations/en.rs and fr.rs as HashMap<&'static str, &'static str>. The selected language is persisted in localStorage (portfolio_language key) and restored on page load.

To add a key:

// en.rs
map.insert("my.key", "My value");

// fr.rs
map.insert("my.key", "Ma valeur");

Usage in a component:

let i18n = use_context::<I18nService>().unwrap();
view! { <p>{move || i18n.t("my.key")}</p> }

---

Notes

• No SSR, no server functions — pure CSR WASM. Dynamic data (news feed) is pre-fetched by GitHub Actions and served as static JSON.
• GitHub Pages source must be set to "GitHub Actions" in repository settings (not "Deploy from branch").

---

License

MIT — see LICENSE.