kacey@ieee.org

Resume

Kacey Wang Software Developer / Applied Scientist

Professional Summary

I'm a creative, empathetic, self-driven developer who uses data to develop intelligent solutions that are intuitive to users and explainable to collaborators

  • > Leverages strengths in adaptability, lateral thinking, and effective communication to quickly integrate into and begin contributing on existing development teams
  • > Fluent in using Tensorflow and PyTorch to build deep learning models in Python, with experience in both training models from scratch and fine-tuning pre-trained models for domain-specific tasks
  • > Skilled at mining high-quality information from heterogeneous data sources and delivering reliable insights by judiciously implementing data-driven machine learning algorithms and domain-driven rule-based methods

Education history

Master of Science in Computer Science University of Southern California Jan 2018- Dec 2019

  • Courses: Deep Learning, Machine Learning, Foundations of Artificial Intelligence, Analysis of Algorithms, Information Retrieval & Web Search Engines, Web Technologies, Database Systems, Multimedia Systems Design

Bachelor of Science in
Neuroscience, Biomedical Engineering
Tulane University Aug 2013 - May 2015

  • magna cum laude

Employment History

Teaching AssistantUSC Department of Computer ScienceJan 2019 - Dec 2019

  • Co-produced the graduate-level, 200+ student course CSCI-572: Information Retrieval & Web Search Engines
  • Topics covered: web crawling; NLP, document parsing, data indexing, query processing; big data ecosystems (Hadoop, MapReduce); sorting and ranking algorithms; Google business logic, SEO, Adsense, Ad Exchange

Project Manager, Research ScientistBJC Institute of Health Jun 2015 - Dec 2017

  • Designed "neuron-on-a-chip" devices for modeling and analyzing complex neural networks in vitro
  • Optimized methods for rapidly fabricating microfluidic devices, achieving feature resolutions down to 10 microns without specialized equipment
  • Mentored interns in designing research experiments, maintaining documentation, and analyzing data

Intern, Associate ResearcherWashington University in St. LouisJun 2010-Jun 2015

ServerMultiple Casual and Fine Dining RestaurantsJun 2010-Aug 2013

  • Thrived working front-of-house for 40-80 hrs/wk, juggling multiple orders and customer requests simultaneously and generating $500-$1000+ in sales per 4-6 hr shift
  • Received the”Above and Beyond” award at the Cheesecake Factory for outstanding guest feedback

Projects

Amazon Review Generator
Deep Generative Models & Transfer Learning in NLP

  • Built a Generative Language Model from scratch with PyTorch that outputs coherent long-form product reviews in response to inputted product keywords and ratings
  • Achieved 49% improvement and alleviated posterior collapse by: (1) incorporating a conditional decoder and hyperparameter-tuned discriminator into the architecture of the Variational AutoEncoder (VAE), and (2) prioritizing data cleanliness
  • Developed a comprehensive data wrangling strategy using PySpark; steps included profiling the metadata and analyzing the text using pre-trained sentiment analysis models to identify poor-quality or fake reviews in a semi-automated manner
  • Presented a more promising solution in supplementary work using transfer learning. Generated realistic reviews with greater semantic and syntactic variety after just 30min of training by using OpenAI's massive pre-trained model (GPT-2 355M) in place of an untrained model

Dynamic Ad Injection
Algorithm-based Feature Detection and Audio Signal Processing

  • Developed a media player that not only transcodes raw digital media files, but also algorithmically replaces any existing commercials based on a brand's presence in the noncommercial content
  • Implemented a fully affine invariant image comparison algorithm (A-SIFT) using OpenCV to match flat graphics to logos appearing in real-world scenes using OpenCV
  • Reduced runtime >450% over baseline by: (1) refining rule-based heuristics to minimize the total number of video frames selected for feature matching without increasing the false negative rate, (2) analyzing the audio signal in the log-power domain to empirically validate an energy drop threshold that precisely identified scene boundaries

Search the LA Times!
Full-text Document Search Engine + Flask App Interface

  • Built a news article search engine app powered by Apache Solr
  • Created a Python-based client API/web UI using Flask that issued HTTP GET requests to the search backend and returned a Google-like search engine results page (SERP)
  • Configured multithreaded Java web crawlers (crawler4j) to fetched ~20,000 unique news articles from the LA Times for populating the search database
  • Simplified mapper and reducer classes and dramatically reduced backend development costs by refactoring Hadoop MapReduce jobs into Spark streaming jobs
  • App features full-text search, auto-correct based on Norvig's spell correction algorithm, data-driven phrase auto-suggest optimized with n-gram tries, optional PageRank scoring (calculated independently using NetworkX)

Weather App x3
RESTful APIs in PHP, JavaScript (Angular + Node.js), and Swift

  • Developed and presented a trio of weather apps built end-to-end using modern web frameworks
  • Created a shared Node.js REST microservice which connects to external web APIs (Dark Sky and Google) to deliver real-time data securely to both web and mobile/iOS interfaces
  • App Features: real-time weather forecasts for inputted addresses or current geolocation; multiple views rendered on demand with interactive charts and widgets for effective data visualizations; saving locations to local storage; social media integrations

https://wang.works (This site)
Django Web App

  • Created a website in Python using Django to show samples of my work and serve as a playground for learning about or experimenting with design patterns and new web frameworks
  • Deployed and managed with git to Heroku; Backend uses a Postgres database for storing unique page content and Google storage buckets to host media and static assets

Skills & Tools



Areas of Interest

  • Machine Learning
  • Deep Learning & AI
  • Data Analysis
  • Information Retrieval
  • Natural Language Processing
  • Web Development
  • UI/UX Design

Technical Skills

  • Unix/OSX, Linux
  • Python
    • matplotlib, numpy, pandas, pyspark, scipy, sklearn
    • tensorflow, keras, pytorch
    • gensim, fasttext, spacy, textblob, nltk
  • JavaScript, Java, MATLAB
  • PHP, Node.js, HTML, CSS
  • Algorithms, Data Structures, Big-O Notation
  • Hadoop/Spark MapReduce, GCP Dataproc
  • Database systems, SQL, noSQL, ERD
    • PostgreSQL, mySQL, Tinkerpop
  • Multimedia systems, digital signal processing
  • RESTful APIs, JSON, XML, micro-services
  • Google Cloud Platform, Heroku
  • Git version control

Languages

  • English (native)
  • Chinese (fluent)
  • Spanish (6 yrs)