Boxian’s Valley

Profile

Boxian Wen
Boxian Wen

Shenzhen, China

Computational Ecology

Biography

I enjoy exploring new things and am interested in practical AI applications.

Background
Master’sUCL · Research Master’s in Computational EcologyBachelor’sSun Yat-sen University · BSc in Marine Biology
Languages
NativeMandarin & CantoneseWorking proficiencyEnglish & Japanese

Research & Practice

Projects

Personal project · Built independentlySynced 2026.11.03

Qilin QILIN

Search & Recommendation System

An integrated system built on the public Rednote Qilin dataset: preprocessing, features, multi-route retrieval, ranking, deployment and online serving, with a Vue + Vite interface.

1,983,938notes
15,482users in dataset
44,024 / 6,192search train / test requests
83,437 / 11,115rec. train / test requests
  1. 01Data
  2. 02Features
  3. 03Retrieval
  4. 04Ranking
  5. 05Serving

Implementation

  • Retrieval: DSSM dual tower + Faiss IVFPQ, Swing and UserCF; lexical / semantic fusion for search.
  • Ranking: LambdaMART (LightGBM + XGBoost) coarse ranking and DIEN sequence-aware fine ranking.
  • Serving: FastAPI + Redis with behavior writeback; Vue + Vite feed, details, recent activity and metrics dashboard.
Pipeline · known-positive retention
ScenarioTest requestsHitRate@10HitRate@50Recall@100
Search6,19255.23%78.42%63.71%
Recommendation11,1152.08%6.61%4.10%
Ranking · complete logged-exposure evaluation
ScenarioTest exposuresNDCG@10AUCGAUC
Search128,1910.54680.58130.5671
Recommendation182,5670.61950.61430.6211

Snapshot: 2026-11-02. All official test requests, without request sampling. A request includes multiple note exposures; these evaluations use different denominators. The test set was previously observed and point-in-time feature lineage remains unverified; results are offline diagnostics.

Serving latency & data sources
Real-time inference latency reported in the repository README
ScenarioP50P95P99
Search223.60 ms496.60 ms654.80 ms
Recommendation228.90 ms369.40 ms425.10 ms

Latency is reported by the project repository and depends on hardware and runtime conditions. README · Full metrics snapshot

Education & Research

Education

University College London UCL

Research Master’s · Computational Ecology

Sun Yat-sen University

B.Sc. · Marine Biology

Research experience

Environmental DNA assessment of coral communities

Zoological Society of London · ZSL

Analysed seawater environmental DNA from the Chagos Archipelago, combining it with CoralNet-assisted image annotation to assess coral composition across lagoon and exposed habitats.

Detected 5 additional scleractinian genera missed by conventional surveys; published as first and corresponding author in Environmental DNA.

Comparative transcriptomics of deep-sea mussels

Southern Marine Science and Engineering Guangdong Laboratory (Zhuhai)

Studied tissue differences in mussels from cold seeps and hydrothermal vents, performing RNA extraction, primer design, qRT-PCR, bioinformatics analysis and validation.

Identified elevated expression of host-defence genes in gill tissue; published as second author in Journal of Oceanology and Limnology.

Contact

Always happy to discuss applied AI, engineering and research.