ILING CHEN

CC Job Tracker — AI-Powered Job HuntCC 求職追蹤器 — AI 輔助求職工具

Full-stack SaaS job tracker with Cloudflare Workers backend. Scrapes 200+ UK remote jobs daily from 6 boards, AI match scoring via Groq (llama-3.3-70b), cover letter generator via Claude, personalised digest to Discord/Telegram/email, and a Chrome MV3 extension for one-click job saving.

全端 SaaS 求職追蹤器，Cloudflare Workers 後端。每日從 6 個平台抓取 200+ 英國遠端職缺，Groq AI 評分、Claude 生成求職信、每日個人化推播，搭配 Chrome MV3 擴充功能一鍵儲存職缺。

CCshop — Digital Services PlatformCCshop — 數位服務電商平台

Firebase-powered e-commerce and account hub for digital services. Handles unified SSO cookie across all subdomains, Stripe billing, full admin CMS (bulk pricing, orders, invoices), VIP tiers, coupon system, and serves as the identity centre for the CC platform ecosystem.

Firebase 驅動的數位服務電商與帳號中樞，整合跨子網域 SSO、Stripe 訂閱付款、後台 CMS（批次定價、訂單、發票）、VIP 方案與折扣碼，為整個 CC 平台生態系提供統一身份驗證。

CC Status — Uptime Monitoring & Incident PipelineCC Status — 全平台監控與事件中樞

Public-facing uptime monitoring SaaS for the CC platform ecosystem — Cloudflare Workers cron probes every 5 minutes, D1 for history, and a live public status page. Also doubles as the ecosystem's shared incident pipeline, receiving webhook reports from every other CC product.

面向大眾的 CC 平台生態系上線監控 SaaS——Cloudflare Workers 排程每 5 分鐘探測一次，D1 儲存歷史紀錄，並提供公開的即時狀態頁。同時兼任整個生態系共用的事件回報中樞，接收其他 CC 產品送來的 webhook 通報。

AI Morning Brief — Autonomous News Digest AgentAI 早報系統 — 全自動新聞摘要 Agent

Fully autonomous daily news agent — pulls from 6 sources (Hacker News, OpenAI/DeepMind blogs, Anthropic, TLDR AI, Reddit), dedupes against history, summarises via Groq with a 70b→8b model fallback, and posts to Discord. State only commits after a successful send, so failures never silently drop articles.

全自動每日新聞 Agent——整合 6 個來源（Hacker News、OpenAI/DeepMind 官方 blog、Anthropic、TLDR AI、Reddit），對照歷史去重，用 Groq 摘要（70b→8b 模型自動降級），推播到 Discord。已讀狀態只在發送成功後才寫入，避免失敗時文章被誤標消失。

Automated Commit Security Scanner自動化 Commit 安全掃描系統

Self-built pre-commit security scanner deployed across 21 production repos. Runs git diff --cached through a headless Claude Code call in safe-mode to catch injected secrets and High/Critical vulnerabilities before they're committed, with automatic installation wired into new repos via a PostToolUse hook.

自製 pre-commit 資安掃描系統，已部署到 21 個正式 repo。每次 commit 前用 safe-mode headless 呼叫 Claude Code 掃描 git diff --cached，攔截外洩密鑰與 High/Critical 漏洞，並透過 PostToolUse hook 讓新專案自動安裝。

Network Intrusion Detection System (IDS)網路入侵偵測系統（IDS）

MSc dissertation at University of Essex — built a full end-to-end, leakage-safe ML pipeline for flow-based network intrusion detection, comparing classical models (Random Forest, SVM, k-NN) against deep architectures (MLP, CNN, LSTM) across 11 attack classes on ~300,000 flow records.

Essex 大學碩士論文——建立完整端對端、防洩漏 ML pipeline，比較傳統模型與深度架構在約 30 萬筆流量記錄及 11 種攻擊類別上的表現。

• Engineered a strict split-then-fit preprocessing pipeline to eliminate data leakage — a critical flaw in most published IDS studies
• 建立嚴格的先分割後擬合預處理 pipeline，消除資料洩漏——多數已發表 IDS 研究的關鍵缺陷
• Compared 3 feature selection families using Jaccard overlap and cross-seed stability; identified 5–10 feature core achieving ≥95% peak macro-F1
• 比較 3 種特徵選取方法，找出 5–10 個特徵核心達到 ≥95% 峰值 macro-F1
• Evaluated ROC-AUC, PR-AUC, Brier score, ECE; applied Platt and isotonic calibration for improved probability reliability
• 評估 ROC-AUC、PR-AUC、Brier 分數和 ECE；套用校準方法提升機率可靠性
• Wilcoxon signed-rank tests confirmed H4: ≤10 features achieves ≥95% peak F1 — clear performance–latency trade-off for SOC deployment
• Wilcoxon 符號秩檢定確認 H4：≤10 個特徵可達 ≥95% 峰值 F1，為 SOC 部署提供清晰的效能–延遲權衡
• Best result: RF + RFIMP (K=5) macro-F1 = 1.000 at 0.43s training; SVM + MI (K=10) F1 = 0.999 sub-second inference
• 最佳結果：RF + RFIMP（K=5）macro-F1 = 1.000，訓練時間 0.43 秒；SVM + MI（K=10）F1 = 0.999