• Designed and implemented a caching service to optimize user matching, achieving monthly cost savings of thousands of dollars. Analyzed service inefficiencies to uncover additional potential savings.• Developed data pipelines utilizing Airflow, Spark (Scala), Python, Kubernetes, Docker, and PostgreSQL, Pandas and implemented APIs using Flask, Flask-RESTPlus, SQLAlchemy, and Kubernetes.• Identified and corrected data inconsistencies in collaboration with data science and product teams.• Extended and deployed IaC (Infrastructure as Code) pipelines for AWS Batch, EKS, and EMR using Ansible and Terraform.• Enhanced system stability with end-to-end tests via Playwright in Jenkins CI/CD, integrated unit testing, and extended Jenkins CI/CD for optimized deployments.• Tech Stack: Airflow, Spark (Scala), Python, Kubernetes, Docker, PostgreSQL, Pandas, Flask, SQLAlchemy, Jenkins, Ansible, Terraform, AWS Batch, S3, EKS, EMR

- Reduced cost of running data pipeline for flood prediction product by 18%. Analyzed GCP instances for kubernetes node pools and selected cheaper options- Root cause and fix CI/CD pipelines, services on Kubernetes, Argo, Docker, Python, Postgres/PostGIS, GCS. Collaborate with Data science, QA, Devops for investigation and testing- Took power network models from proof of concept in Jupyter notebooks to data pipelines in production using Argo, snowflake, kubernetes, docker, python, pandas, pytest- Identified inputs for Flood and Seismic prediction product and then extended them to new geographical locations. Fixed critical bugs affecting data quality. Used Postgres with PostGIS, Python, Kubernetes, Argo, QGIS, Docker, GCP- Built system to calibrate Flooding model for large geographical areas. Collaborated closely with Data Scientists. Used Python Multiprocessing, Kubernetes and Docker to reduce runtime by half when compared to the initial version- Reduced test runtime in local and CI/CD by 35% using pytest-xdist and improved developer experience on Python repo

- Slashed API times by 13% or 0.5s by pushing computations for AWS rekognition data closer to the database using embedded SQL. Also saved >$10k on AWS redshift cluster. Used scala, javascript and datadog for monitoring- Reduced costs by 85% for media pipeline which can process more than 11k media per day by migrating from hive logo to google logo service while maintaining precision and reducing recall. Used AWS lambda step functions on python- Enhanced 2 million media with color tags. Researched, deployed and optimized color detection model as a part of media pipeline on AWS fargate to keep error rate under <1%- Adressed data duplication for >1k accounts and ∼100 million Impressions by integrating Universal App Campaign data from google ads API in our ad performance ETL pipeline scala, java and datadog for monitorig- Extended ability to understand media’s compliance with platform best practices by adding new media source into computer vision model. Projected revenue in the first year was $600K. Used AWS dynamoDB and lambda- Deployed NLP model for quantifying ad fitness. Used Python, Docker and Flask with AWS step functions and lambda- Quantified ad fitness by deploying NLP model with AWS step functions and lambda. Used Python, Docker and Flask- Provided the ability to get granular data of top creative elements in an ad. Integrated feature split across 3 pipelines in scala, python and javascript into single python ETL pipeline from data stored in AWS redshift and MySQL- Improved test coverage by 8% over 6 months by pushing unit-testing with mockito for our scala application- Examined impact of iOS14+ changes on platform integrations with facebook, adwords, snapchat, twitter, tiktok. Collobarated cross functionally with vidmob’s data and business unit to understand and communicate impact- Expanded snapchat ad ETL pipeline with multi asset story ad data in Scala

Assisted study of child tantrum psychology by developing WearOS smart watch app in java. Added ability to store and retreive Accelerometer, Gyroscope and Heartrate data

- Architected a technical framework for MIT spinoff company to visualize and make sense of usage data from their connected wearable product worn by more than 10K users- Defined metrics and visualized them to expose the performance of the product and the mobile app- Designed and implemented data pipelines using python, pandas, REST APIs and SQL- Built a dashboard in Google Data Studio from data stored in BigQuery and other sources- Identified previously undetected logging faults in app usage logs. Corrected them using SQL and API calls- Automated pipelines using python scripts on google cloud instance and google app engine- Performed data visualizations using R and ggplot in jupyter

Structured multiple 1500+ line unstructured R scripts into easy to use package using Devtools and Roxygen

- Awarded for excellent execution on a Web Dashboard and data pipeline for search and display ads, and app store data- Created ETLs on petabyte-sized datasets from search, app logs on Microsoft Cosmos using Scope, powershell and SSIS- Automated tasks using SQL, powershell, R and VBA scripts. Saved more than 30 man-hours for my client with a single script- Built a sentiment analysis model to quantify app performance. Used naive bayes and maxent on scraped app-store comments- Forecasted advertisers performance in the holiday season using triple exponential smoothening and other methods- Quantified the effect of attributes like category on app downloads and revenue using hypothesis testing(ANOVA)