Hallucination DetectionDesigned and built prototype for use in PoC project with clientPerformed experiment comparing solutions across detection strategies, models, and application tasksBiomarker PredictionPredicted sequence variants that predict therapeutic outcomes for drugs using the link prediction modelImplemented second level model based on foundation graph model to improve predictionWrote paper based on approach submitted for peer-reviewCustom Knowledge GraphsBuilt knowledge graph for retail domain, and incorporated data from multiple popular product ontologies Built knowledge graph for industrial chemical domain, and incorporated sales data to create user recommendationsBillion-scale Biomedical Knowledge GraphBuilt multiple dashboard from graph data for clientsBuilt and deployed link prediction models on the knowledge graph using DGL, AWS Redis, and AWS LambdaBuilt ETL Pipeline for data from Wikidata, Clinical Trials, NIH Grants using Apache Spark and AWS NeptuneManaged multiple junior data scientists in building natural language graph query featureCOVID-19 Knowledge GraphBuilt experimental knowledge graph from CORD-19 dataExperimented with Link Prediction model to expand partner’s QSAR GraphCheminformatics modelingBuilt experimentation notebooks, defined and implemented metrics.Led team of six UW Chemical Engineering students in building small ADMET modelsMentored junior data scientist on project

Embedded on Indeed Questions team● The team’s products are the job detail questions (“What is the sales medium?”) asked of employers, and screener questions (“How many years of driving experience do you have?”) asked of applicants.● Built a Bayesian inference model of the probability that a given screener question type will be selected. Implemented so new types could be incorporated without re-deployment.● Researched effects of the product on user behavior in order to create a team metric. Joined months’ worth of data to estimate the effect our product has on application rates, and quality of applications● Mentored three data scientists, and defined data science projects, and worked with product managers to maintain data science backlog● Owned team data, data quality, and data engineering● Analyzed 6 years of job descriptions to assist with government research projectPresentations & Education● Gave five tutorials on the basics of Apache Spark and mentored multiple data scientist and software engineers across the organization in using Apache Spark for ETL, data analysis, and machine learning tasks.● Gave lectures on NLP, including basics of NLP, survey of NLP Libraries, vocabulary analysis.● Consulted regularly with multiple teams on data science, NLP, data engineering, and troubleshooting problems with Apache Spark, Apache Pig, and in-house data tools.

● Gave six tutorials on the basics of Apache Spark and Databricks to teams across the organization, as well as tutored teammates, and performed notebook reviews● Led team in building tool to measure transcription accuracy of speech-to-text models; the tool took sound files from S3, processed, and measured transcription accuracy via Databricks● Built ETL pipelines on Databricks for music data gathering that scraped data from multiple sources, cleaned and joined with fuzzy string matching, then saving in redshift, and CKAN (data catalog)

Natural Language Processing Library (Aug 2016 – Oct 2016)• Built NLP library: interfaces, annotators, local and Spark-based pipelines• Built resource abstraction library• Built and maintained maven-based jobs on Jenkins, and project repository on Git

● Built Oozie workflow that loaded data from MySQL, transformed, re-encrypted data, and produced configurable data quality summaries. This pipeline was used on terabytes of personal financial data.● Built tool for cross-referencing US Census data and Melissa data which was used to improve fairness in financial credit monitoring and fraud detection models● Set up Jenkins and ELK (elasticsearch, logstash, kibana) for data science organization, and gave lecture on the importance of software engineering quality and software development process in data science

Clinical Analytics (Sep 2014 – Sep 2015)● Built data quality analysis report pipeline using Apache Hive. It consumed configuration to check that data was adhering to business logic, as well as generating basic data metrics like size, type, null counts● Built models for hospital patient readmission prediction in multiple projects on terabytes of data, as well as working with customer service team to present and explain results in a dashboard● Designed NLP library that used Apache Spark with OpenNLP and a library of clinical NLP functionsSearch Engine for TREC 2014 Clinical Decision Support track (Jun 2014 – Aug 2014)● Team’s goal was to build a search solution to find relevant biomedical articles for patient case reports.● Served as SCRUM master for two interns and maintained shared experimentation environment. Clinical Auto-Coding (Apr 2013 – May 2014)● The team worked on predicting assigned codes using multi-label models● Built text analysis components, featurization DSL, and experimentation pipeline.● Maintained experimentation environment, and worked with ops team to improve its securitySearch Engine for TREC 2012 Medical Records track (Jun 2012 – Aug 2012)● Built a search solution for cohort selection, our results in top three for preferred metric● Researched prior results and created list of best practices for experimentation strategy

Clinical Auto-Coding (Apr 2013 – May 2014)• Built text analysis components, featurization DSL, and experiment pipeline. Used UIMA, Spark.• Staged data and maintained experimentation environment.Hospital Patient Readmission Prediction and Fraud Detection (Nov 2012 – Mar 2013)• Performed data quality analysis and statistical analysis for a 9 TB national dataset. Used Hive.xRelevance (proprietary search engine) (Sep 2012 – Oct 2012)• Defined and implemented search performance metrics; improved NLP corpus processing.

Search Engine for TREC 2012 Medical Records track• Identified best practices from TREC 2011 Medical Records Track.• Built experimentation pipeline based on Indri search engine. Used Solr, Indri• Presented for Atigeo’s team at TREC.

Services for Students with Disabilities• Facilitating test-taking for students with physical, mental, and emotional disabilities• Assisting individual students, as well as groups of students with homework and studying in mathematics, computer basics, logic, and natural science• Supplementing lessons for students with learning disabilities• Test preparation for groups of students