Ayushi Agarwal Email & Phone Number

Mountain View, California, United States

San Francisco, Ca, Us

Chime is a neo bank leader in the U.S. banking segment. The mission of the company is to democratize financial service and help its members lead healthier financial lives by avoiding bank feesData Science (DS) and Machine Learning(ML) PlatformDeveloped ML Platform to empower DS run end-end ML lifecycle for large scale ML models, starting from experimentation, model development & deployment. -Auto Retrain Framework: Led development of model auto retrain framework to mitigate ML model degradations thus saving ~60 million/year from model refreshes. Partnered with cross functional teams, PM’s to define roadmap, technical design and did impact analysis to prioritize refresh initiatives . Executed development of components like pipeline orchestration, evaluation framework (shadow/production, offline/online) and deployment automation to increase the model release velocity by 12x thus reducing cross functional dependencies- DS Exploration Environment: Led development of secure, compute/storage scalable workspace to develop and tune large scale models with the production data, spearheaded the brainstorming sessions, TDD; configured a new AWS environment using terraform with services & resource permissions dedicated for the ML team. Using python, CI/CD built tooling to containerize, schedule and deploy ML workflows across different AWS environments. Onboarded 20 data scientist, and several ML cross partner teams to the development platform thus increasing the development productivity by 10x- ML Real time Orchestration: Developed a self-serve real time inference framework for 20+ production services with strict latency and availability SLAs of 5000s of features in sub-300ms to increase DS velocity. Proactively ensured operational excellence of services - defined contracts, monitors (data dog, ML observability) for customer teams to deliver on performance, availability, and reliability commitments. Performed PR reviews and organized guild sessions for technical excellence

San Jose, Ca, Us

1. AI/ML Platfrom Engineering:- Design and developed distributed machine learning platform for data scientist, engineers and analysts on both CPU and GPUs using Kubernetes.- Design and developed framework to simplify/automate the model development and serving through Kubernetes. Used docker to containerize 40+ kernel images and containerized notebooks to orchestrate running of kernel pods using Enterprise Gateway and CICD pipeline. This helped in decoupling of pods from local server and successful scaling of the platform to high compute GPU environment. Presented idea in JupyterCon 20202. PayPal Notebooks: - Using Python, developed & open sourced Jupyter based Elastic Notebooks – a central data exploration and DS/engineering platform for PP. Developed code for custom magics, utilities, enabled Github, tableau integration and incorporated ML pipelining capabilities using Airflow for analyticsThis one-stop solution provided PayPal’s data science community a unified way to access and analyze data from platforms like Spark, Hive, HBase, Kafka, Teradata, & SQL ultimately onboarding of 6500+ customers on the platformPresented in Teradata Universe Data conference and Women in Data Science 20193. Spark Compute as a Service (ScaaS): -- Using Scala, incorporated multi-version support capabilities (Apache Livy open source contribution), Hbase Support, and SQL compatibility into Livy compute platform- This enabled successful porting of 1 million Spark applications through Rest API, reducing operation overhead of maintaining different configurations4. Gimel: Unified compute/data platform to run big data apps on multiple compute engines and data storages. (HDFS, HBASE, ElasticSearch,Kafka(Streaming),MongoDB) with SQL support : https://gimel.readthedocs.io/5. BOT Platform: ‪Using Python and BERT transfer learning, developed a chatbot for help-slack channels thus reducing RTB effort for 4500 + customers across PayPal and increase engineering efficiency by 85%

San Jose, California, Us

Provide data harmonization, product performance measurement, and descriptive/predictive analytics to Finance sector companies- Using MS SQL, created fact tables to successfully standardize business data from various platforms. This ultimately helped the client aggregate dashboard reports and visualize operational and marketing metrics across various categories like geography, default rate, etc. - Using Python, implemented business/data quality rules on source data to perform validation and eliminate 97% of data discrepancies

San Jose, Ca, Us

Built predictive model for estimating target environment’s resource availability and forecasting run time performance of 65000 BTEQ scripts, that needed migration from Teradata to Hadoop Data Lake - Prepared parquet format test data on Hadoop and wrote Java UDFs to convert BTEQ SQLs to Hive SQls- Coded and executed several Spark streaming jobs using Scala and captured respective performance logs from Kibana- Constructed training file and used H20 Sparkling Water to build predictive model using GLM, GBM machine learning algorithms. The model predicted target environment’s total CPU time and potential resource consumption with 85% accuracy

Marina Del Rey, California, Us

Project Patent Infringement:Overview of the project:Building Knowledge Graphs”on Patent Monetization Entities to find similarities in previous fraudulent case and current judicial case. Mine huge amount of cases to find patterns in cases filed and outputs of those cases and use that feature to identify PMEs- Worked with Dr. Knoblock and Dr. Szekely on information crawling, information integration and on building knowledge graphs of Patent Monetization Entities-Crawled public legal websites and patents and used Python and Hadoop Map Reduce to develop ontology extraction models using "Web Karma"Project 'Pegasus’ Workflow Management System: - Used data mining and visualization techniques through R programming language to evaluate and analyze deviations in observed scientific workflow performances across various end-to-end performance models - Used a combination of offline and online strategies and developed a analytical/predictive model for detecting performance anomalies during workflow execution using machine learning algorithmsLink: http://pegasus.isi.edu

Marina Del Rey, California, Us

Brand Sentiment Analysis of Apple Watch and Apple Pay:Project Scope: The project is to measure the brand sentiment on Apple watch and Apple Pay from real world data(Twitter hash tags) using Map-Reduce Hadoop Framework(implemented in Python)Goal: To review sentiments, product features, product comparisons e.g. Fit bit surge, Samsung Gear SBy tapping this information, market analysts and Apple R&D team can get a good idea of the consumer sentiment, pros and cons of their product and the features the customer expects, product competitors and use this data to determine what needs to be incorporated in their next product release.Execution:Data Source: Twitter (Streaming API) Data Mining: PythonServer to Hold Big Data​: Amazon EC2 ServicesFramework Implemented: Hadoop (Manual Configuration using Linux Servers)Map-Reduce scripts: Python Data Visualization: Python

Bangalore, Karnataka, In

Part of the software engineering team that built and enhanced ‘Conexus’ – a specialty pharmacy management system for US retailers- As the Senior Software Developer in 8-member team, handled all SDLC phases including domain driven design, coding, testing, deployment and post deployment troubleshooting. Project entailed programming in C++, Unix Shell scripts, and PL/SQL- Awarded ‘Star of Quarter’ for taking sole ownership of enhancing key components in insurance management module using C++, that resulted in 50% reduction in bug fix requests and significant improvement in process efficiency- Developed efficient and re-usable UNIX shell scripts to perform mass updates/data retrievals using SQL and PL/SQL in Informix and Teradata databases for more than 5000 production sites- Established an efficient training model to impart technical training to fresh graduates, resulting in shortened transition time to work

Master Of Science (M.S.), Data Informatics

Bachelor Of Technology (B.Tech.), Electrical Engineering Technologies/Technicians