Uppsala, Sweden

Wenyu

NLP / AI Engineer · Uppsala University

  • AI Detection
  • Croatian News Information Retrieval
  • Explainability (IG/LIME/ATTENTION) on Transformers
  • Multilingual MT (mBART / NLLB)
GitHub LinkedIn CV Email
Quick demo
$ python -m venv .venv && source .venv/bin/activate
$ pip install -r requirements.txt
$ python train.py --model bert-base-uncased --task sst2
$ python explain.py --method ig --k 10
✔ saved: outputs/ig_heatmap.png

Highlights

Beyond the code.

Cross-cultural & multilingual collaboration

Comfortable working in international teams. Fluent in English and Chinese, with an academic background in Urdu. I communicate clearly across cultures and time zones.

Reliable, structured, and easy to work with

I work in a structured way: define scope, document decisions, share progress early, and deliver reliably. I ask good questions and keep stakeholders aligned.

Tech stack I use most

Focused on NLP and machine learning (Transformers/BERT-style models), with an emphasis on evaluation and error analysis. I enjoy building baselines and iterating with experiments.

My tech stack

PythonPyTorchHugging Facescikit-learnTransformersRAGVector DBFastAPIDockerGitLinuxSQL

Projects

A few things I built recently.

RAG retrieval engine web interface

RAG Retrieval Engine

Built a retrieval pipeline with BM25/dense/hybrid search , reranking, and evaluation for robust, explainable RAG performance.

RAGFAISSHybrid search
AI-generated content detection web interface

AI-Generated Text Detection (Internship Project)

Built an AI-generated text detection pipeline for short-form content, covering data prep, modeling, evaluation, and deployment-ready inference.

DetectionNLPML
Information retrieval evaluation metrics (Accuracy@5, MAP, etc.)

Croatian News Information Retrieval

This repository evaluates ranking algorithms for a Croatian-language IR system using a news-article setup.

BM25SBERTEvaluation
Deletion test plot comparing IG / Attention / LIME

XAI for Text Classification

Integrated Gradients vs LIME/SHAP with deletion tests on fine-tuned BERT (SST-2).

XAIBERTCaptum
BLEU scores comparison before and after fine-tuning

Low-Resource MT: Luganda ↔ English

Training + back-translation experiments with mBART/NLLB; BLEU + qualitative error analysis.

MTLow-resourceNLLB

About

It's me.

I am a Master’s student in Language Technology at Uppsala University, specializing in Machine Learning and Natural Language Processing (NLP). My academic background combines linguistics and computer science, with hands-on experience in text classification, sentiment analysis, information extraction, and transformer-based models (BERT, GPT, etc.). I am skilled in Python, PyTorch, Hugging Face, and scikit-learn, and have experience working with large-scale datasets and applied AI projects. Proficient in English, Chinese, and Urdu, I thrive in collaborative, international team environments.

Contact

Let’s build something together!

Best way to reach me: email.