andreipaciurca.github.io

Live: andreipaciurca.github.io

A terminal-styled, ATS-optimized resume site with a fully automated LinkedIn → Gemini AI → GitHub Pages data pipeline. Written in TypeScript, compiled to native ES Modules — no framework, no bundler.

Features

• Automated pipeline — GitHub Actions fetches LinkedIn data via Apify, summarizes it with Google Gemini Flash (latest), and redeploys on every push or monthly cron.
• Curated fallback content — hand-crafted experience bullets and summary are used when AI output fails quality checks, so the site never shows raw data.
• Terminal UI — modular TypeScript with a CLI aesthetic, 3D profile photo flip (Braille ASCII art), and typewriter animations.
• ATS-optimized print layout — dedicated print layout with automatic overflow detection and a heuristic quality score (0–100); Harvard style, single-column, Arial font.
• Dark / Light mode — CSS variable–driven theming, no JS involved for the color switch.
• Type-safe codebase — full TypeScript strict mode, shared interfaces in js/modules/types.ts, declaration file for the auto-generated data module.

Architecture

.github/workflows/
  update-resume.yml       # Triggers: push to master, monthly cron, manual dispatch

assets/
  style.css               # All site styles — theme variables, layout, print rules
  favicon.png             # Site icon
  profile-photo.jpg       # Profile photo (used for ASCII art generation + OG image)

scripts/
  update-resume.js        # Pipeline: LinkedIn JSON → Gemini → profile-data.js

js/
  main.ts                 # Application entry point, event wiring
  modules/
    types.ts              # Shared TypeScript interfaces (ProfileData, Experience, AtsReport…)
    config.ts             # UI text constants, print strategy steps
    state.ts              # AppState interface + animation handle tracking
    dom.ts                # Typed DOM element references (HTMLButtonElement, etc.)
    utils.ts              # Pure utility functions (escape, truncate, URL validation)
    ui.ts                 # Animations, Braille ASCII art generation, typewriter
    renderer.ts           # Pure HTML string builders for each content section
    print.ts              # ATS print layout, overflow strategy, heuristic scoring

tests/
  resume.spec.js          # Jest unit tests (56 tests across 6 suites)
  e2e/
    site.spec.js          # Playwright E2E (Chromium, WebKit, Firefox)

data/
  linkedin.json           # Cleaned LinkedIn export (auto-synced by pipeline)

health/
  index.html              # Human-readable pipeline health page

profile-data.js           # Single source of truth — all resume content (auto-generated)
profile-data.d.ts         # TypeScript declaration for profile-data.js
health.json               # Pipeline run status — served as JSON API endpoint
tsconfig.json             # TypeScript: ES2020 target, strict, in-place compilation

Pipeline

push / cron / dispatch
        │
        ▼
  actions/checkout@v4
        │
        ▼
  Apify: harvestapi~linkedin-profile-scraper
  input: {"queries": ["https://www.linkedin.com/in/andreipaciurca/"]}
        │
        ▼
  update-resume.js
  ├── strip moreProfiles, receivedRecommendations, photo fields
  ├── Gemini Flash (gemini-flash-latest alias) → summary + bullets
  ├── quality gate: looksLikeRawDump() + bulletsAreGood() — rejects bad AI output
  ├── fallback to FALLBACK_SUMMARY + FALLBACK_EXPERIENCES when AI fails
  └── write profile-data.js + health.json + cleaned data/linkedin.json
        │
        ▼
  git commit (amend if last commit is already chore: sync profile data)
        │
        ▼
  force-push → GitHub Pages redeploy

The amend-on-cron strategy keeps history flat — repeated automated runs update the same commit rather than adding new ones.

TypeScript

All js/** modules are TypeScript. The compiled .js files are committed alongside the source so GitHub Pages can serve them without a build step.

npm run build       # compile .ts → .js (tsc, in-place)
npm run typecheck   # type-check without emitting files

Key TypeScript decisions:

• "moduleResolution": "bundler" — preserves .js import extensions required by browser native ES modules.
• "rootDir": "js" + "outDir": "js" — in-place compilation, index.html unchanged.
• "strict": true + exactOptionalPropertyTypes — catches undefined/null mismatches at compile time.
• profile-data.d.ts provides types for the auto-generated profile-data.js without modifying the pipeline output.

Setup

Required GitHub Secrets

Secret	Purpose
APIFY_TOKEN	Apify API token for the LinkedIn Profile Scraper
GEMINI_API_KEY	Google AI Studio API key (must have no HTTP referrer restrictions)

Configure at Settings → Secrets and variables → Actions.

Note: The GEMINI_API_KEY must be unrestricted by referrer — the pipeline runs server-side on GitHub Actions and cannot send a browser Referer header.

Local Development

ES Modules require a local HTTP server — opening index.html directly will not work.

# Serve the site
python3 -m http.server 8080
# → http://localhost:8080

# Run the data pipeline locally
GEMINI_API_KEY=<key> node scripts/update-resume.js

# Type-check
npm run typecheck

# Compile TypeScript
npm run build

Testing

npm test              # Jest unit tests (56 tests, 6 suites)
npm run test:e2e      # Playwright E2E — Chromium, WebKit, Firefox
npm run test:all      # both suites

Unit test suites:

• Profile Data — Structure — required keys, types, contact validation
• Profile Data — Content Quality — summary format, bullet constraints
• Profile Data — Skills Integrity — icons, duplicates, whitespace
• Health Endpoint — schema, freshness (< 90 days old)
• Source File Integrity — DOM IDs, pipeline patterns, CSS, workflow
• TypeScript Setup — tsconfig, compiled output, interface exports, scripts

Health

Pipeline status is available at andreipaciurca.github.io/health/.

Indicator	Meaning
[ai] gemini	Gemini produced usable AI output this run
[sync] apify	Fresh LinkedIn data fetched this run