● Full-time SDE working on asynchronous lambda invocations traffic, where requests are enqueued and executed such that the client does not need to wait for a response. ● Projects o Added the AsyncEventsReceived metric, which communicated to customers how many invocations were successfully received and enqueued, providing them a greater understanding of the system.o Modified system infrastructure that allowed routing messages to an increased number of availability zones, improving fault tolerance and reducing fleet size by 20%. o Added detection logic to locate runaway functions, functions that loop infinitely, inside asynchronous destinations.o Built infrastructure and expanded production service into three new regions: Zurich, Hyderabad, and Melbourne. ● Operationso Five week-long on-call duties over a six month period. Solved over 70 urgent tickets and 100 total tickets on a global scale. Responsibilities included mitigating urgent situations in production regions, root-causing significant events such as networking failures, responding to customer inquiries, and clearing backlogged items.o Completed operational and maintenance tasks to improve availability, security, reliability and performance of the system.o Actively engaged in design reviews and provided thoughtful feedback.o Assign and prioritize tasks during sprint planning.● Toolso Amazon Elastic Kubernetes Service (EKS), AWS CloudFormation, AWS CloudWatch, Java, DynamoDb, AWS Lambda, AWS Certificate Manager (SSL/TLS certificates), AWS Brazil (Internal Artifactory tool), Safe Dynamic Config (Internal configuration management automation tool), AWS Route 53 (Scalable DNS system)

Used a combination of sample-based MPC and cutting edge instance segmentation algorithms, this work focused on balancing a tray with a ball using a robotic arm. Sample-based MPC was utilized due to its ability to respect constraints, optimize a cost function, uses a system’s dynamics model, and learn from actively taken samples. Real-time feedback such as position, velocity, and acceleration was provided through computer perception of the ball. Integrating this perception with sample-based MPC provided more enriching feedback and the ability to solve more complex robotic manipulation problems.Thesis: https://drive.google.com/file/d/1E9Zz1b2xT3Yy5Y82E0uER63cJ5LDAJ-N/view?usp=share_link

The current approach for resource allocation for a customer is static. While this works, it is very inefficient since some customers will be assigned far too many resources and others will be assigned too few. I built a dynamic algorithm to properly assign the correct amount of resources. This saves AWS Lambda significant amounts of resources and improves the customer experience.

Background:Citation analysis is the examination of the frequency, patterns, and graphs of citations in documents. Citation count is the number of times a paper has been cited or referenced by other papers. Citation count is very important in Academia since it determines a paper's impact and as a result, funding distribution, author's impact on the field, and promotion/tenure. However, citation count is an unreliable metric since it can be influenced by various factors such as institution, journal, or field. Thus, I sought out to examine citation trends from millions of papers to find patterns that may be able to better describe a paper's quality.Procedures:• Created a querying pipeline that used APIs from Elsevier, PUBMED, and NIH to acquire citation data for millions of papers• Created a pipeline to organize and clean citation data and create a standard database• Visually analyzed patterns by reducing high dimensional data to low dimensional data and clustering.• Fit various functions to citation data to determine which best fit the data• Created a machine learning classifier to classify the shape of citation data• Determined if clusters changed over time• Examined correlations between cluster membership and various factors such as journal, institution, author etc.Results:• Successfully created a pipeline that wrangled citation data for four million papers• Successfully organized, cleaned, and filtered citation data• Successfully reduced high dimensional data into low dimensional data with 95% data retention • Successfully clustered the citation data into five unique clusters• Successfully fit numerous logistic, linear, and polynomial functions to the citation data. Determined 5th degree polynomials generally fit the best• Successfully created a machine learning classifier that could classify paper shape with a 86% accuracy• Currently determining cluster change and correlations

Background: An ovarian germ cell teratoma is a tumor made up of several different types of tissue, such as hair, muscle, or bone arising from gamete cells. Little is known about the genomics of immature ovarian teratomas. Thus, I sought out to conduct a copy number analysis on ovarian teratomas and compare it to other ovarian germ cell tumors in order to shed more light on the origination of ovarian teratomas.Procedure:• Created reads from DNA sequences of tumor and normal biopsy genomes• Processed read by mapping them to a reference genome and outputted them to an analysis-ready file• Performed bias correction• Calculated tumor fraction and copy number events via Hidden Markov Models• Streamlined the process to enable faster manipulations and error detectionResults:• Successfully generated and processed reads.• Successfully applied bias correction to enable analysis.• Calculated tumor fraction for all samples.• Found a some samples returned a tumor fraction of 0 with others being greater than 0.25.• Created copy number graphs that showed genetic amplification for numerous samples.• Successfully streamlined the process such that the entire pipeline would execute with one call.

• Teaching CSE190B: Direct Admit Seminar• Creating lesson plans about UW Research• Providing fun activities to get my students passionate about research• Bringing in guest speakers to inform my students about UW research

• Lunar tubes are tunnels on the moon that could one day act as temporary shelter for humans. • I sought to develop scripts for a robot to identify minerals in a lunar tube.• In addition, I wanted my machine learning framework to be customizable, easy-to-use, and portable.• Created an image classifier using supervised machine learning to detect objects in static images.• Utilized TensorFlow to classify various rock and minerals in one-go.• Simplified user-entered parameters in order to easily interchange training databases.• Also, programmed in Google CoLab to allow cross-platform execution of the scripts.• Finally, provided documentation to allow more advanced users greater tuning of the machine learning model.

Background:Sensory hair cells, critical for hearing, are vulnerable to multiple damaging agents, such as noise and drugs. In humans and other mammals, which lack regenerative abilities, hair cell loss leads to permanent hearing impairment. In contrast, non-mammalian vertebrates exhibit robust regeneration. I set out to characterize amino acid-level differences between regenerating and non-regenerating species, thereby identifying similar protein sequences in regenerators that are not present or altered in mammals and may contribute to differential regenerative capacity. Procedures:• Retrieved all protein sequences for the zebrafish genes from the NCBI database• Determined genes from other species that evolved from the same common ancestral gene as the zebrafish genes and retrieve their respective proteins• Processed and cleaned the data to prepare it for analysis• Reduced potential bottlenecks for the analysis• Analyzed the proteomic data by comparing regenerator and non-regenerator species to determine significant differences• Confirmed if the significant proteins contribute to hair cell regeneration in regenerator speciesResults:• Analysis identified proteins with highest similarity between non-mammalian species (threshold = > 0.8) and high differences between mammalian species (threshold = < 0.2)• Proteins included inositol hexakisphosphate kinase 2a, chondroitin sulfate glucoonyltransferase, NP_001107909, collage type XXVIII, catechol O-methyltransferase, transforming acidic coiled-coil-containing protein 3, armadillo repeat protein, cortex-3, TERF-1, E3 ubiquitin-protein ligase SHPRH.• Successfully located the gene, determined target sequence, ran a PCR and gel, created gRNA and Cas9 Protein for Terf-1