About half-way through my time at SCSIO I was referred to also work at ABC, it was here where I learned about building an AI-powered chatbot for CFA exam preparation, focusing on the backend infrastructure and intelligent question-answering system.
designed a LangChain RAG pipeline with GPT-4o to process 1.5K+ CFA exam questions for context-aware Q&A generation
developed a RAGAS evaluation framework measuring faithfulness, answer relevancy, and context precision across 10 test sets, enabling systematic prompt engineering improving accuracy from 68% to 88%
created a auto-classification system using TF-IDF and cosine similarity to automatically tag questions to 100+ knowledge points, reducing manual labeling time by 20+ hours per week
Software Engineering Intern
South China Sea Institute of Oceanology, Guangzhou, China | Python
May - August 2025
Developed a computer vision system to predict embryo viability (fish and chicken eggs) using hyperspectral imaging and machine learning. Working closely with marine biology researchers to build and deploy models to automate embryo quality assessment.
Forever grateful to be able to work on a machine learning stack so early in my career.
collaborated with researchers to implement a PyTorch training pipeline with HRNet architecture scaling data processing to 10K+ chicken and fish embryo samples for viability analysis with automated feature extraction
improved embryo viability prediction accuracy from 72\% to 83\% by optimizing XGBoost and Random Forest classifiers with 5-fold stratified cross-validation
deployed a Flask API for researchers with request batching for on-demand embryo viability predictions with sub-2s latency
analyzed feature importance and spectral sensitivity to identify key predictive wavelength bands, enhancing interpretability and model optimization
Built a mobile-first educational platform designed to help teachers support students with autism.
Focused on improving the user interface and backend systems to make the application faster and more maintainable. It was this experience that made me realize
just how important it is to a product to have a appealing functional user interface.
refactored SQL queries into a promise-based ORM, improving asynchronous processing and maintainability
worked alongside designers to implement the redesigned mobile-first UI with lazy-loading and optimized routing, reducing initial app load time by ~35% and eliminating navigation lag on Android devices
consolidated MVC-style login and registration controllers into a single controller, reducing code duplication by 38%