Portfolio

My Github Blog

May 1st, 2024 — I made a promise to rebuild from the ground up.
After two years studying data science, I realized I could use algorithms but didn’t understand them deeply enough.
So I started over — posting every day to rebuild my intuition and technical depth.

Over a year, I revisited Statistics, Probability, Machine Learning, and Databases, then expanded into Deep Learning, LLMs, and MLOps. That journey transformed me — from learning algorithms to leading real-world ML & MLOps projects.

Through this commitment, I’ve written over 200 in-depth posts exploring everything from the foundations of statistics to the architecture of transformers and production-grade ML systems.

Each post became a reflection of growth — documenting not only what I learned but how I built, deployed, and optimized real systems.
Today, the blog stands as a living record of my continuous evolution from student to Machine Learning & MLOps Engineer — grounded in curiosity, consistency, and craftsmanship.

🔗 Visit My Blog →
📄 Download My 200 Days Challenge Summary (PDF)

Data Science Project

[LLMOps — LLM-Powered Real-Time Translation System]

Designed and implemented a real-time multilingual translation system that combines advanced AI with scalable cloud infrastructure to process and translate continuous news streams with ultra-low latency.

🧩 Architecture & Key Components

Amazon EC2 (7 instances): Multi-server setup for distributed scalability
Kafka: High-throughput message streaming backbone
FastAPI: Asynchronous API for low-latency translation requests
Hugging Face (Helsinki-NLP/opus-mt-es-en): Spanish → English translation model
Docker: Ensures consistent, portable deployments
Prometheus + Grafana: Real-time monitoring and system observability

🚀 Impact

Achieved sub-second translation latency under real-time workloads
Built a fault-tolerant multilingual pipeline ready for production deployment
Delivered a fully observable system tracking latency, throughput, and translation quality in real time

🔗 Project Resources

🔗 Real-Time Tweet Sentiment Analysis Pipeline

Built a real-time tweet sentiment classification pipeline on Databricks using Spark Structured Streaming and Transformer models, with MLflow for experiment tracking and Delta Lake for fault-tolerant storage.
Delivered live dashboards visualizing sentiment trends across millions of tweets.

Analyzed 500K+ multilingual social media posts (English & Spanish) to uncover public perceptions of e-cigarettes using BERTopic and RoBERTa, identifying cross-lingual sentiment shifts and key discussion themes.

🔗 Cross-Cultural NLP Analysis of Luxury Hotel Reviews in Europe — LDA Topic Modeling

| LDA | Topic Modeling | NLP | Sentiment Analysis | Cross-Cultural Analytics | Python | Visualization |

Explored European luxury hotel reviews using LDA topic modeling to uncover country-specific satisfaction drivers, linking linguistic tone to cultural preferences and customer experience.

Developed an interactive dashboard comparing education, income, housing, and healthcare metrics across New York, New Jersey, and Connecticut to evaluate regional quality of life patterns.

Performed multivariate statistical modeling on global video game sales to uncover genre, platform, and regional trends, providing insight into market dynamics and sales forecasting.

COVID-19 Awareness & Prediction Analysis through ML & DL

Developed ML/DL models to predict COVID-19 trends in Ohio using feature engineering and time-series modeling, revealing key behavioral and social factors influencing public awareness.

Resume and Other Docs

🙌 Most Updated Resume Here

Email Me

Wonha Shin / leahnote01@gmail.com 📩 Email me!