Algorithm Engineer focused on Search, Recommendation, Ranking, and Intelligent Backend Systems

我当前以算法工程师为主线，重点聚焦搜索、推荐、排序、特征工程与智能后端系统。研究背景主要来自 UCL 计算生态学研究型硕士与 中山大学 海洋生物学本科训练，并在环境 DNA、转录组、生信分析与 AI 辅助珊瑚评估中积累了分析经验。

• 求职方向：算法工程师 / 搜索推荐 / 智能后端
• 并行兴趣：AI / 多模态 / 珊瑚保护 / 计算生态学 全奖博士机会
• 当前代表项目：Rednote Qilin Search & Recommendation System

Rednote Qilin Search & Recommendation System · 11/2025 – 05/2026
受小红书启发的搜索推荐一体化个人项目，独立完成数据预处理、特征工程、多路召回、粗排、精排、线上部署与在线服务的全流程闭环。

• 技术栈：Python FastAPI Vue3 TypeScript Redis DuckDB Faiss LightGBM XGBoost DSSM DIEN
• 数据规模：约 200 万 笔记、1.5 万 用户、10 万 推荐样本、5 万 搜索样本
• 模型架构：DSSM 双塔 + Faiss、Swing、UserCF 多路召回；粗排采用 LambdaMART (LightGBM / XGBoost)；精排采用 DIEN
• 在线服务：基于 FastAPI + Redis，支持实时行为写回与搜推联动，线上延迟稳定在 ~150ms
• 离线指标：
  • 搜索侧：HitRate@500 = 0.88，Recall@500 = 0.65，MRR@100 = 0.11，NDCG@10 = 0.69，AUC = 0.77
  • 推荐侧：HitRate@500 = 0.99，Recall@500 = 0.99，NDCG@10 = 0.87，AUC = 0.84

Repository

I am primarily targeting algorithm engineering roles, with a focus on search, recommendation, ranking, feature engineering, and intelligent backend systems. My background combines a research master's in Computational Ecology at UCL with marine biology training at Sun Yat-sen University, along with hands-on work in environmental DNA, transcriptomics, bioinformatics, and AI-assisted coral analysis.

• Main direction: Algorithm Engineer / Search & Recommendation / Intelligent Backend
• Parallel interest: fully funded PhD opportunities in AI, multimodal learning, coral conservation, and computational ecology
• Current flagship build: Rednote Qilin Search & Recommendation System

Rednote Qilin Search & Recommendation System · 11/2025 – 05/2026
An integrated search-and-recommendation build inspired by Xiaohongshu, covering data preprocessing, feature engineering, multi-channel retrieval, coarse ranking, final ranking, deployment, and online serving.

• Stack: Python FastAPI Vue3 TypeScript Redis DuckDB Faiss LightGBM XGBoost DSSM DIEN
• Data scale: about 2M notes, 15k users, 100k recommendation samples, and 50k search samples
• Modeling: DSSM dual tower + Faiss, Swing, and UserCF for retrieval; LambdaMART (LightGBM / XGBoost) for coarse ranking; DIEN for sequence-aware final ranking
• Online serving: FastAPI + Redis, with real-time behavior writeback and linked search-recommendation serving, stable latency around ~150ms
• Offline metrics:
  • Search: HitRate@500 = 0.88, Recall@500 = 0.65, MRR@100 = 0.11, NDCG@10 = 0.69, AUC = 0.77
  • Recommendation: HitRate@500 = 0.99, Recall@500 = 0.99, NDCG@10 = 0.87, AUC = 0.84

Repository