Hi, I'm Yeyang Liu

AI Researcher & ML Engineer & Kaggle Competition Expert

Master's student at Universität Zürich, specializing in Artificial Intelligence and enjoying taking on challenges. Looking for opportunities in AI-driven brain decoding & BCI.

View Projects Get in Touch
👩🏻‍💻

// About Me

Background and technical expertise

🎓 Education

M.Sc. Informatics — Universität Zürich (5.4/6)
Major: Artificial Intelligence | Minor: Data Science Related courses: Machine Learning for NLP · Model Analysis and Interpretability in NLP · Deep Learning · Deep Generative Models · Prompt Engineering B.Sc. Computer Science (AI) — Universiti Malaya (3.95/4) Related courses: Machine Learning · Natural Language Processing

🏆 Achievements

Kaggle Competition Expert Silver Medal — NeuroGolf 2026 (29th / 2,963) Silver Medal — AI Mathematical Olympiad: Progress Prize 3 Bronze Medal — Learning Equality Curriculum Recommendations Top 5% (community) — LLM Agentic (Swiss) Legal Information Retrieval CLEF Shared Task 1st Place — GutBrainIE 2025 & 2026 (NER · Entity Linking · Relation Extraction) 2nd Place — TalentCLEF 2026 Task B (Job-Skill Retrieval · Ranking)

// Skills

Click a skill to filter the projects below.

Python NLP LLMs RAG Information Retrieval Embeddings Vector Databases FAISS Prompt Engineering Hugging Face vLLM Transformer Knowledge Graphs Graph Neural Networks Anomaly Detection PyTorch Geometric Recommender Systems PyTorch Pandas Signal Processing AutoGluon LightGBM Git/GitHub Scikit-Learn Linux TensorFlow MATLAB

// Projects

Selected works and research

2026 🏆 2nd · 9th · 3rd

Mobile Health Activity Monitoring

Built task-specific PPG and IMU pipelines for breathing rate, heart rate, and activity monitoring, placing 2nd / 47, 9th / 47, and 3rd / 46 across three course challenges.

PPG IMU Signal Processing AutoGluon LightGBM PyTorch
View Details → 2026 GSoC 2026 · GENIE

GENIE Graph Anomaly Detection

Built an unsupervised graph autoencoder pipeline for LHCO collider events. Adjacent-subjet regularization raised BB1 AUROC from 0.891 to 0.923 and MaxSIC from 2.40 to 3.87.

Graph Autoencoder PyTorch Geometric FastJet LHCO Topology Regularization
View Details → 2026 🥇 1st Place

GutBrainIE 2026 System

Information-extraction system for gut-brain axis PubMed abstracts, covering named entity recognition, entity linking, and relation extraction. Ranked 1st in all four GutBrainIE 2026 subtasks.

BiomedBERT CRF SapBERT Relation Extraction FAISS
View Details → 2026 🥈 2nd Place

TalentCLEF Job-Skill Retrieval

Retrieval system matching free-text job titles to the most relevant skills in the ESCO taxonomy. Reached 0.7913 graded nDCG for 2nd place, pairing a fine-tuned bi-encoder with LLM reranking.

JobBERT GIST HyDE LLM Reranker vLLM ESCO
View Details → 2026 ⚖️ Top 5% · Kaggle

Swiss Legal RAG

Retrieves the Swiss federal law citations a Bundesgericht decision would cite for an English legal question. Bilingual retrieval + LLM reranking + LangGraph agentic verification; ranked top 5% on Kaggle.

Agentic Retrieval BGE-M3 Qwen2.5-32B vLLM DeepSeek LangGraph Co-citation Graph
View Details → 2026 Datathon

RobinReal Challenge

Conversational multi-modal property search across four languages, treating listing photos as a first-class retrieval signal. Built the hybrid retrieval stack fusing dense, sparse, and cross-modal image rankers.

BGE-M3 SigLIP-2 ONNX + FAISS RRF MCP
View Details → 2025 Master Project

MR Anomaly Detection

Deep learning-based anomaly detection for MR-only radiotherapy QA. Benchmarked 10 models using Anomalib, achieved 0.90 patient-level Dice with PatchCore and validated multi-center generalizability.

Anomalib PatchCore Medical AI NIfTI
View Details → 2025 🥇 1st Place

Biomedical NER System

Developed a MedicalBERT-CRF system for extracting entities from gut-brain PubMed abstracts. Improved micro-f1 by 3% using ensemble and data augmentation strategies.

BERT CRF NER BioNLP
View Details → 2024 Knowledge Graph

Graph-Based Movie Chatbot

QA chatbot answering natural language questions over a movie knowledge graph. Integrated BERT, fuzzy matching, and LLM-based NER with hybrid SPARQL and embedding-based retrieval.

SPARQL BERT LLM KG
View Details → 2024 LLM + RAG

Virtual Patient System

Medical education chatbot using Llama2 and RAG. Features a Virtual Patient module simulating patient interactions and an MCQ generator using MilvusDB and Dragon+ model.

Llama2 RAG MilvusDB Medical AI
View Details → 2023 🥉 Bronze Medal

Curriculum Recommender

Kaggle competition solution matching curriculum contents with topics. Used SimCSE for retrieval with iterative hard example training, achieving 76th place.

SimCSE Transformers Ranking
View Details →

// Small Tools

Selected utility projects

Gmail → Telegram

Personal Gmail Assistant

Automatically scan and filter Gmail messages with predefined keywords, generate AI summaries based on contents, and send a daily digest via Telegram.

Python Gmail API Telegram Bot LLM

// Experience

Professional background

Algorithm Developer Intern

MISUMI Group Inc.
Jul 2023 – Sep 2023 | Shanghai, China
  • Developed domain-specific QA chatbot using ChatGLM2, comparing LoRA fine-tuning with RAG (Milvus vector DB)
  • Improved item-based recommendation from word2vec to widedeep, increasing ROC by ~3%

Algorithm Developer Intern

MISUMI Group Inc.
Aug 2022 – Jan 2023 | Shanghai, China
  • Applied ML for user segmentation and profiling aligned with business logic
  • Refactored recommendation project, switching from deepwalk to lightfm
  • Developed price prediction model using AutoGluon with CAD-extracted features

// Get in Touch

I'm always open to discussing AI research, collaboration opportunities, or interesting projects.

Email LinkedIn GitHub Kaggle