• Created Prediction system to estimate arrival times for multimodal shipments. Led and developed most components of the system which comprised of: • Clustering to identify locations (dbscan, and custom variants of it to stop chain reaction/cluster expansion based on sensor dwell time strength and other domain features to better discover target locations automatically) • Graph/Networkx to identify routes between shipment locations and using it to forecast future paths based on past preference+distance hybrid heuristicBoosted Trees (lightgbm) with domain+auto-discovered cluster features to predict dwell times at locations. • Recurrent Neural Networks(RNN)/Transformers to predict transition times Well engineering architecture to allow multiple parts of the system to swap in with various models (e.g. swapping Dwell predictor with statistical predictor when we have low amount of data) • Custom crowdsource/3rd party integrate to assimilate output of Ocean+Air legs/schedules into the overall prediction framework.• Created supply chain loops identification system. • Created custom algorithm to Identify repeating trips of Assets through various POIs • Example Loop: Discovery of container type asset’s movement through retail locations to Analyze and Predict Turn Around Time (TAT)• Created ETL Data Pipelines using Apache Beam (Python SDK) which used custom processing steps. •Used ElasticSearch, Mysql, Cassandra, and Snowflake to assimilate and process various data about IOT Sensors + Assets + Shipments •Created an advanced pipeline to merge data from multiple sources and apply clustering to discover location and use discovered locations to further discover visitation loops.

• Created Bluetooth based auto-inventory tracking solution, which used historical Rssi data and past inventory to automatically map retail/sold zones in customer premises and used those statistical models at runtime to Estimate Inventory at retail shelves vs in Sold zones• Experimented on ETL pipelines using Dagster, Apache Spark, Apache Beam, Flink.Created Anomaly tagging system using Gaussian Mixture Model to tag live sensor feed of gps/gsm points with • Created Interactive analysis tools using holoviz Panel framework, which enabled us to rapidly productionize ad-hoc analytics work.• Worked on Node.js backend where I worked on FIFO queues (AWS Sqs, Redis queue, Kafka) and created data processing code which horizontally parallelizes operations per sensor to make the whole solution scalable.• Worked on Chatbot (using Rasa tensorflow embedding pipeline) chatbot to create internal diagnostic tool

• Helped migrate backend written in Java Spring Framework to Node.js microservice based backend.• Learned angular.js and created responsive UI for product pages and admin system management pages.• Created distributed web scraper to create internal tools to reduce day to day mundane tasks using chrome extensions toolkit with node.js backend with sockets. • Helped complete the Android application by working on the API consumption layer of the android app.

• Was part of the Innovation wing in TCS which experimented and created POCs.• Worked on business process automation project which was created in java using apache camel framework & spring.

• Android Custom Component Development; Created custom timeline calendar component which used projections to shrink all year events in a single screen where nearby events were given more space and farther events were pushed to horizons.