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。

CC API Platform — Developer Tools HubCC API 平台 — 開發者工具中心

Subscription-based developer tools platform on Cloudflare Workers + D1 + KV. QR code generator, URL shortener with analytics, OG image API, API key management, usage tracking, rate limiting, and cross-subdomain SSO via Firebase.

基於 Cloudflare Workers + D1 + KV 的訂閱制開發者工具平台。QR 碼生成、短網址含分析、OG 圖片 API、API 金鑰管理、用量追蹤，與 CCshop 共用 Firebase SSO。

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 CC Job Tracker and CC API Platform.

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，為所有 CC 平台提供統一身份驗證。

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

EV Charging Behaviour Analysis & Prediction電動車充電行為分析與預測

Analysed a simulated dataset of EV charging sessions to uncover behavioural patterns and predict both charging cost and energy consumption — combining unsupervised clustering with supervised regression models.

分析模擬的電動車充電資料集，揭示行為模式並預測充電費用與耗電量，結合無監督聚類與監督式迴歸模型。

• Cleaned and preprocessed the dataset; EDA and visualisations using Pandas, Matplotlib, and Seaborn
• 清洗並預處理資料集；使用 Pandas、Matplotlib 和 Seaborn 執行 EDA 與視覺化
• Applied KMeans clustering to segment EV usage patterns by charging behaviour
• 應用 KMeans 聚類依充電行為區隔電動車使用模式
• Built Linear Regression and Decision Tree models to forecast charging cost and energy consumed (kWh)
• 建立線性迴歸與決策樹模型，預測充電費用與耗電量（kWh）

Crime & Weather Correlation — Colchester犯罪與天氣相關性分析 — Colchester

Explored crime data in Colchester and its correlation with weather patterns using R, Leaflet, and Plotly. Created interactive maps and visual insights to support local policing strategies.

使用 R、Leaflet 和 Plotly 探索 Colchester 犯罪資料與天氣型態的相關性，建立互動式地圖與視覺洞察，支援地方警務策略。