Jiaqi F.

Key Skills

Software

Other

Top Subject Matter

Intelligent education assistant

RAG and LLM fine-tuning

Medical data science for diabetes risk prediction

Top Data Types

Audio

Image

Text

Top Task Types

Fine-tuning

Data Collection

Classification

Freelancer Overview

Intelligent Assistant System for Smart Education (LLM fine-tuning + RAG). Core strengths include DeepSpeed, WSL2, and LangChain. Education includes Master of Science, University of Nottingham (2025) and Bachelor of Science, Ningbo University (2024). AI-training focus includes data types such as Computer Code and Programming and labeling workflows including Fine-tuning, Evaluation, and Rating.

Labeling Experience

League of Legends Match Outcome Prediction (data collection + feature engineering)

Data Collection

Collected and engineered structured sports match data for outcome prediction by building Python data collection scripts and creating a cleaned training dataset. Used Riot API collection with rate limiting, breakpoint resuming, and error handling to obtain high-rank match records. Performed feature engineering and trained multiple models with tuning and cross-validation. • Collected 5,000+ high-rank match records from the Riot API with robust data ingestion controls. • Cleaned raw data, processed outliers, and engineered early-game difference features. • Split data 80/20 and applied cross-validation plus GridSearch for hyperparameter tuning. • Trained and compared eight models including Logistic Regression, SVM, Random Forest, XGBoost, and MLP.

2025 - 2025

Diabetes Prediction Project (model training and evaluation on tabular medical dataset)

Built classification models for diabetes prediction using a large CDC dataset, focusing on data preprocessing and model evaluation. Performed missing value processing, outlier detection, and cleaning to ready structured tabular inputs for training. Applied class-imbalance optimization and assessed performance using accuracy and ROC-AUC. • Prepared the CDC Diabetes Health Indicators dataset with missing-value handling and outlier processing. • Trained and compared multiple classifiers including Logistic Regression, Random Forest, XGBoost, KNN, and MLP. • Used GridSearch and cross-validation for systematic hyperparameter tuning. • Achieved strong test performance with Random Forest (88% accuracy, 0.96 ROC-AUC).

2025 - 2025

Intelligent Assistant System for Smart Education (LLM fine-tuning + RAG)

Fine-tuning

Developed and fine-tuned large language models locally to support an intelligent education assistant with retrieval-augmented question answering. Focused on preparing LLM inference and RAG pipelines rather than manual data labeling, including document embedding and similarity search for retrieval. Implemented interactive teaching modes combining Q&A and virtual teacher experiences. • Used LoRA fine-tuning workflows with DeepSpeed and local deployment on WSL2. • Built RAG using Chatbot and LangChain, including document embedding and similarity search. • Integrated a MetaHuman virtual teacher and speech recognition with Chinese text-to-speech in UE5. • Delivered a live virtual teacher demonstration using OBS.

2023 - 2024

Education

Work History