Developed low-level Scala APIs and high-level Python APIs to enable large-scale GIS queries with GeoSpark, using quad-tree partitioning. Achieved >5x faster speed than PostGIS queries on PostgreSQL database. This has helped scale the enrichment of geospatial features, such as reverse geocoding, segmenting a trip based on zones, etc on millions of trips. Geo data is a primary feature in insurance scoring. - Leveraged the geo-platform above to refactor and scale an existent GeoPandas pipeline to predict stop-signs on roads in the USA based on GPS trails near road intersections. - Scraped geographic information like boundaries and roads of the entire world from OpenStreetMaps, converted them into Scala-compatible formats, and designed a hierarchical storage to enable the geoplatform APIs. - Automated large dataset generation and validation tasks - processing '00s of millions of rows - on Airflow. - Migrating in-house libraries to be Python 3 compatible to enable using newer machine-learning frameworks.

Integrated a front-end dashboard with NoSQL database using Javascript to display a variety of analytical metrics in near real-time for a popular news-provider in West Bengal. SELECT PROJECTS Visualizing the schema of complex Python objects for easy analysis of their data-structures. Uploaded on PyPI (package name: print-schema) with 1.5K downloads (as of Mar '20). Created a reusable pipeline which fetches top posts from Reddit, creates images from texts, then uploads them on an Instagram page in a completely automated way through APIs. Blogpost featured on Better Programming. Quantifying biases (or the lack thereof) of news sources on Twitter by fetching tweets, filtering them, and per- forming sentiment analysis, implemented with Python and VADER.