👋 About Me

I am currently a first-year Master's student in Computer Science at the Institut Polytechnique de Paris (École Polytechnique & Télécom Paris), a leading research university in France. I specialize in machine learning and artificial intelligence, with a strong interest in reinforcement learning and agent-based systems, particularly for sequential decision-making in real-world interactive environments.

My research experience spans NLP, large language models, recommendation systems, retrieval-augmented generation (RAG), and generative models for domain-specific tasks. I have solid technical skills in Python, PyTorch, and the HuggingFace ecosystem, with experience in designing, fine-tuning, and evaluating LLM-based systems.

I am seeking a six-month internship starting April 1st, 2026, to apply my skills to research on the transition from offline to online learning for computer user agents.

🔥 News

  • 2026.02 Looking for a 6-month research internship starting April 2026 in RL / Agent-based AI.
  • 2026.01 Completed research on Agent-based Text-to-SQL with Memory-Augmented Exploration at Télécom Paris.
  • 2025.09 Started M.S. in Computer Science at Institut Polytechnique de Paris.
  • 2025.08 Completed AI Research internship at Postal Savings Bank of China, working on LLM-based recommendation and diffusion models for NER.
  • 2024.06 Graduated with B.S. in Statistics from Guangdong University of Foreign Studies (Top 5%, GPA: 18.5/20).
  • 2024.06 Received Outstanding Graduate Scholarship (Top 1%).

📝 Publications & Research

Télécom Paris 2026
Text-to-SQL Research

Agent-based Text-to-SQL with Memory-Augmented Exploration

Kaiying Wu

Master Research Project, Télécom Paris, 2026

Designed an execution-aware agent for Text-to-SQL with iterative generate–execute–refine loops. Introduced memory-augmented exploration to store and reuse execution feedback, reducing redundant exploration. Improved execution correctness on Spider 2.0: +1.71 points over a strong agent-based baseline (ReFoRCE) with Qwen-32B.

PDF Code
PSBC 2025
LLM Recommendation

LLM-based Generative Recommendation for Financial Applications

Kaiying Wu

Postal Savings Bank of China, AI Research Internship, 2025

Investigated the applicability of LLM-based generative recommendation for financial product recommendation under strong business and risk constraints. Built an offline next-item prediction framework and achieved a consistent ~6.5% improvement in Recall@K over strong discriminative baseline (SASRec).

PDF
PSBC 2025
Diffusion NER

Diffusion Models for Financial Named Entity Recognition

Kaiying Wu

Postal Savings Bank of China, AI Research Internship, 2025

Applied diffusion-based sequence modeling to financial NER to address noise, annotation inconsistency, and ambiguous entity boundaries. Achieved a 2.87-point improvement in entity-level F1 on the ChFinAnn Chinese financial NER benchmark. Validated on proprietary financial text data.

PDF
Patent
BERT GEC

BERT-Based Unsupervised Grammar Error Correction for Low-Resource Languages

Kaiying Wu

Guangzhou Key Laboratory of Multilingual Intelligence Processing, Patent, 2023–2024

Proposed a BERT-based unsupervised approach for grammatical error correction (GEC) in low-resource languages by reformulating GEC as a multi-class classification problem. Achieved relative macro F-score improvements (91.58% on Tagalog; ~27% on Indonesian) over IndoGEC baseline.

PDF Code
IP Paris 2025
RAG Medical QA

Entity-driven Retrieval-Augmented Generation for Medical QA (MedMCQA)

Kaiying Wu

Institut Polytechnique de Paris, Master Project, 2025

Designed an entity-driven RAG pipeline for medical exam question answering. Built an end-to-end RAG system with dense vector retrieval over Wikipedia-derived documents and explicit fallback to LLM-only inference. Observed consistent accuracy gains (~2–3%) on the MedMCQA benchmark.

PDF Code

💻 Selected Projects

Object Tracking

Object Tracking with Siamese FC & OSTrack

Realized object tracking using Siamese Fully-Convolutional Networks and OSTrack for robust real-time visual object tracking. Implemented and compared classical and transformer-based tracking architectures.

Code Docs
Land-Cover Classification

Land-Cover Classification with ResNet-18 & ViT

Developed land-cover classification models using ResNet-18 and Vision Transformer (ViT) for satellite imagery analysis. Compared CNN and Transformer architectures for remote sensing tasks.

Code Docs
XAI & Representation Learning

Explainable AI & Neural Representation Learning

Explored explainable AI methods and neural representation learning techniques. Additional projects include clinical term normalization and sentiment analysis.

Code Paper

📖 Education

09/2025 – 09/2027 (expected)

M.S. in Computer Science

Institut Polytechnique de Paris (École Polytechnique & Télécom Paris), Palaiseau, France

Coursework: NLP, Computer Vision, LLM, Reinforcement Learning, Agentic AI, Biomedical AI

09/2020 – 06/2024

B.S. in Statistics

School of Mathematics and Statistics, Guangdong University of Foreign Studies, Guangzhou, China

GPA: 18.5/20 (Top 5%)

Outstanding Graduate Scholarship (Top 1%)  |  First-Class Undergraduate Scholarship (Top 5%, 4 times)

🛠️ Skills

Programming

Python, SQL, Bash, Git, LaTeX

AI & ML

Machine Learning, Deep Learning, NLP, LLMs, RAG, Agentic AI, LoRA Fine-tuning, Prompt Engineering

Frameworks

PyTorch, TensorFlow, Scikit-learn, HuggingFace, Pandas, NumPy, OpenCV

Tools

Linux, Jupyter, VS Code, Google Colab

Languages

English (Proficient), Mandarin (Native), Cantonese (Native)