All-in-One

A full-stack, serverless microservices personal web platform that I designed, built, and run in production for near-zero cost (~$0.02/month). It bundles several small products into a single codebase: a portfolio, an English-learning toolkit, a resource/blog manager, an AI stock & gold analysis dashboard, and a codebase-understanding RAG engine (my graduation project).

🌐 Live demo: https://all-in-one-cpp.pages.dev — the portfolio home page is viewable without logging in.

This is my playground for practicing a real product end-to-end: from UI/UX design, building the backend, provisioning infrastructure as code, CI/CD, all the way to cost optimization and security. Each feature solves a concrete problem and demonstrates a distinct skill set.

💰 Deliberate cost optimization: the serverless architecture + Cloudflare free tier keep the real cost at ~$0.02/month, with an AWS Budget ($1/month) and an automatic hard-stop that blocks any overrun.

The home page / is viewable without logging in: hero, about-me, certificates (lightbox), GitHub projects (responsive grid), and a login CTA.

• The intro text comes from profile.bio — the owner edits it in the dashboard and the public page syncs automatically.
• A real-time weather theme (OpenWeatherMap, HCMC time): 5 conditions × 6 time-of-day periods = 30 gradients; the banner renders rain/fog/clouds/thunder/stars effects with SVG + Framer Motion.
• The landing uses one continuous dark canvas + an "aurora" transition zone for seamless section handoff.

• Study schedule by TOEIC/IELTS, per skill.
• Generate tests from PDF: upload a PDF → extract text → build an exam; upload audio (MP3/WAV) for listening sections.
• Vocabulary flashcards using the SM-2 spaced-repetition algorithm.
• Reminder emails sent automatically via SES + EventBridge.

The /topics area holds resource files + blog posts; each topic is public/private. Files upload to R2 via presigned URLs. Real-world problems handled along the way: SPA fallback for dynamic routes on a static export (Cloudflare Pages), and UTF-8 charset for Vietnamese files.

The /stocks dashboard analyzes QNS, VIC (VN equities), GOLD (XAU/USD), and SJC (domestic gold):

• A pure Node pipeline driven by Claude Code Routine (no extra infrastructure): keyless price fetching, technical indicators (SMA, Bollinger, RSI), and a ledger + self-scoring of predictions.
• Interactive SVG charts (crosshair tooltip, progressive line drawing) and an investment-simulation profit/loss panel.
• The full dataset (ledger.json) is being collected to train an offline ML model in Phase 2.

The /codemind engine indexes this very repo and answers questions about it:

• A feature map in the style of Runeterra/LoL: each capability is a "region", routes = dependencies; clicking opens a function-to-function relationship map by role (zoom/pan/pin).
• Two-mode RAG Q&A (general user / developer), with hybrid retrieval BM25 + semantic embeddings (RRF) + feature routing (top-1: 35% → 80%).
• A Cross-Modal Bridge linking docs ↔ code + stale-documentation (drift) detection.
• No dependency on commercial LLM APIs — embeddings/generation run on open-source models via Cloudflare Workers AI (zero-cost) or local Ollama.

_(The job-search feature — job-scraper — is hidden from the UI because the scrape sources are blocked; the backend remains intact.)_

                    ┌─────────────────────────────┐
   Cloudflare Pages │   Next.js 14 (static export) │
   (CDN, free)      │   Portfolio · English ·      │
                    │   Topics · Stocks · CodeMind │
                    └──────────────┬───────────────┘
                                   │ HTTPS + JWT cookie
                    ┌──────────────▼───────────────┐
   AWS API Gateway  │  HTTP API (1 URL, path-route) │
                    └──────────────┬───────────────┘
        ┌──────────────────────────┼──────────────────────────┐
        ▼            ▼              ▼              ▼            ▼
   auth-service  profile-svc  english-svc   topics-svc   notification-svc
   (Lambda)      (Lambda)     (Lambda)      (Lambda)     (Lambda)   + job-scraper
        └──────────────┬──────────────┬──────────────┬───────────┘
                       ▼              ▼              ▼
                  DynamoDB    Cloudflare R2       AWS SES
                  (NoSQL)     (file storage)    (email)   ◄── EventBridge cron

• Frontend: Cloudflare Pages (output: 'export', no SSR).
• Backend: 6 NestJS Lambdas — auth, profile, english, topics, notification, job-scraper.
• Storage: DynamoDB (data) + Cloudflare R2 (files, zero egress).
• No Lambda Function URLs (blocked by an SCP) → routing goes through API Gateway.

Details: PROJECT_CONTEXT.md · doc/architecture.md · doc/deployment.md.

All-in-One/
├── frontend/              # Next.js 14 App Router (public + private routes)
├── services/             # 6 NestJS microservices (each deploys to 1 Lambda)
│   ├── auth-service/         # JWT auth (no signup, seeded admins)
│   ├── profile-service/      # Portfolio, certificates, GitHub repos
│   ├── english-service/      # Schedule, PDF→tests, SM-2 vocabulary
│   ├── topics-service/       # Resource files + blog (R2 presigned)
│   ├── notification-service/ # Reminder emails (SES)
│   └── job-scraper-service/  # Job scraping (hidden from UI)
├── shared/               # Shared types & constants
├── infrastructure/       # Terraform IaC (all AWS resources)
├── stock-analysis/       # Stock/gold analysis pipeline (Claude Routine)
├── codemind/             # RAG indexing pipeline + Cloudflare Worker
├── doc/                  # Detailed per-feature documentation
└── docker-compose.yml    # Local dev (DynamoDB + LocalStack + MailHog)

• Full-stack TypeScript — Next.js (App Router, static export) + NestJS.
• Serverless & microservices architecture — service separation, API Gateway, Lambda packaging.
• Infrastructure as Code — fully modularized AWS infrastructure with Terraform.
• CI/CD — GitHub Actions running test/build per changed unit, automated deploys, secret scanning.
• Cloud cost optimization — near-$0 architecture + Budget hard-stop.
• Security — cross-site JWT, CORS with credentials, prompt-injection guard for the RAG, no committed secrets.
• Applied AI/ML — RAG (BM25 + vector + rerank), open-source embeddings, time-series analysis.
• Frontend craft — SVG animation, dynamic weather theme, anti-slop design system.
• Testing — ~465 tests (Jest backend, Vitest frontend, node --test for pipelines).

# Requirements: Node 20+, pnpm 9+, Docker, AWS CLI v2, Terraform 1.7+
pnpm install
docker-compose up -d        # DynamoDB Local + LocalStack S3 + MailHog
pnpm dev                    # Frontend at http://localhost:3000

# Checks before pushing
cd frontend && npx tsc --noEmit && pnpm build
pnpm -r run test            # backend/shared (Jest)
cd frontend && npx vitest run   # frontend (Vitest)

_Built end-to-end with Next.js · NestJS · AWS Serverless · Cloudflare — by Nguyễn Phúc Thọ._