Cassie Guo | Data Scientist

Hi, I'm Cassie Guo

A

Promote the well-being of people via data and models

About

I dissect data, distill insights, develop solutions and deploy applications for problems of all sizes and shapes. My data science philosophy includes:

Data Centric

No Free Lunch

Occam's Razor

The best way to come up with the utimate idea is to have lots of ideas

Projects

                
Medical Research Interpreter
                  An multi-stage reasoning application to summarize and interpret medical literature
                
AccomplishmentsTech stacks:python, Langchain, huggingface, LLM, Prompt Engineering
Using large language models (LLMs) to summarize medical literature, and translate into layman language
Stage 1: A large language model to summarize literature with short sentences
Stage 2: A different model using text2text generation to translate the summary into layman language for non-professionals

Explainable API to Detect Public Health Misinformation
                  On Twitter
                  
                  An API which flags misinformation on social media using ensemble BERT
                
AccomplishmentsTools: Python, HuggingFace
API end point that uses ensemble BERT fine-tuned on COVID-19 related tweets to classify new tweets 
Augmented dataset to incorporate domain expert knowledge

Exploring Impact Factors for BTC Fee using Regression Models
                  Using open source datasets to understand factors that influence BTC transaction fee
                
AccomplishmentsTools: Python, regression models

X-Ray Image ClassificationDetect COVID-19 from Chest X-ray Images using Transfer Learning                
                
AccomplishmentsAchieved 98% sensitivity and 99% specificity scores
Implemented one cycle learning policy and reduced learning rate policy on ResNet50, resulting in better performance within shorter training time

Kaggle competition - M5 Forecasting Accuracy
                  Use the daily sales of each goods from Walmart during 2016-2019 to predict the daily sales of next 28 days
                
AccomplishmentsTech stack: time series forecasting, GBM, feature engineering 
Selected features, and incorpated lags and moving windows on time series data
Final solution is a catboost model which achieved 16% score on public leaderboard

A draft implementation of pricing agent in DRL
                  Using openAI gym to model a pricing agent with deep reinforcement learning (DRL)
                
AccomplishmentsExperimented PPO, A2C and several other policies using stable baseline library

Conversion Rate Prediction by Two-layered Stacking Model + GPU
                  Supercharge conversion rate prediction on large volume itineraries
                
AccomplishmentsTech stack: python, parallel computing, GPU, pyspark
Two-layered ensemble model to predict itineraries for search ranking

Contact

xg233dataforthewin@gmail.com

github.com/lilsummer

linkedin.com/in/xi-cassie-guo-datascience