Machine Learning Platform- Worked on developing a lite version Flipkart’s ML Platform capable to deploying ML model as REST API servicesSearch Semantics- Worked on remodeling Flipkart’s Search Semantic system, for externalistion, responsible for understanding user query to provide most suitable search resultsAPI Gateway- Worked on developing an API Gateway service, using Spring Cloud framework, as a single point of entry for multiple backend… Show more Machine Learning Platform- Worked on developing a lite version Flipkart’s ML Platform capable to deploying ML model as REST API servicesSearch Semantics- Worked on remodeling Flipkart’s Search Semantic system, for externalistion, responsible for understanding user query to provide most suitable search resultsAPI Gateway- Worked on developing an API Gateway service, using Spring Cloud framework, as a single point of entry for multiple backend services.- Implemented AuthN & AuthZ following OAuth2.0 with route level rate‐limiting and circuit breaking capabilities.StoreFront ‐ Flipkart in a box- Worked on developing Storefront, a module for FK in a box to serve targetted content on HomePage of a marketplace.- Adopted and reused existing services with high availability requirements using Guava cache and Embedded‐Solr for low latency.- Worked on creating a near real time Spark streaming data pipeline, deployed on Dataproc (GCP) cluster, taking inputs from various streaming sources such as Kafka. Used various datastores to achieve different objectives like Aerospike as intermediate datastore and Hbase as the primary datastore.Reporting Framework- Developed a service to provide real‐time responses on top of large datasets for complex analytical queries needed by various reports on UI dashboard, using Clickhouse as data‐warehouse.- Built a data pipeline to transfer data from primary datastore(MySQL) to Clickhouse using Debezium and Kafka ConnectAssortment Intelligence- Developed a service to provide real‐time responses on top of large datasets for complex analytical queries needed by various reports on UI dashboard, using Clickhouse as data‐warehouse.- Built a data pipeline to transfer data from primary datastore (MySQL) to Clickhouse using Debezium and Kafka Connect Show less

Product Matching and optimisation- Optimised Product matching algorithm using parallelisation which reduced run time from 7 days to 5 hours (98% reduction) and decreased memory utilisation by 75%.- Implemented image similarity search in product match algorithm which improved match rate by 3400%.- Developed and trained an Autoencoder (a type of DNN) to convert images to embeddings, for Vector Indexing using FAISS- Used Sentence Transformers to convert text embeddings and… Show more Product Matching and optimisation- Optimised Product matching algorithm using parallelisation which reduced run time from 7 days to 5 hours (98% reduction) and decreased memory utilisation by 75%.- Implemented image similarity search in product match algorithm which improved match rate by 3400%.- Developed and trained an Autoencoder (a type of DNN) to convert images to embeddings, for Vector Indexing using FAISS- Used Sentence Transformers to convert text embeddings and built a Vector Index on top of it for textual similarity / semantic search.Web Scraping- Revamped the backend component for web scraping by rebuilding it from the ground up using Playwright and BeautifulSoup. Improved testing and debugging capabilities of web crawlers and transformed the component into an independent service.- Designed the LLD for web crawlers while adhering to OOP principles of polymorphism and inheritance. Implemented the Factory pattern to enable the creation of crawlers with varying functionalities.Competitive Intelligence for brands- Took ownership for repurposing our existing product, Competitive Intelligence, for brands, aimed at giving insights about their competitiveness with other brands across different channels- Designed and implemented an end-to-end pipeline with robust data ingestion, web scraping setup for generating boiler plate code for web crawlers, customised report generation for clients and used Azure infrastructure for seamless deployment.- Spearheaded the redesign of the existing crawling system, leveraging advanced technologies such as RabbitMQ and Celery to enable large-scale operations while implementing robust tracking and monitoring features for multiple tenants. Show less

Porting all the C++ APIs of InnerModel Library in Robocomp framework to Python3 using numpy, scipy and pybullet.

-> Worked on Multimodal Semantic Segmentation for Self Driving Car Applications. -> Woked on training DL models on 3D point clouds for semantic segmentation and re-projecting the result on a 2D image.

• Implemented an Image Processing Algorithm (Image Fusion) to enhance the degraded underwater images in real-time before feeding it to the perception module of the vehicle.• Integrated UUV Simulator, an open-source underwater simulator, to work with AUV-IITK code base• Designed and developed the software architecture for AUV consisting of dedicated layers for hardware integration, controls & navigation, motion planning, and perception worked on developing software for Autonomous… Show more • Implemented an Image Processing Algorithm (Image Fusion) to enhance the degraded underwater images in real-time before feeding it to the perception module of the vehicle.• Integrated UUV Simulator, an open-source underwater simulator, to work with AUV-IITK code base• Designed and developed the software architecture for AUV consisting of dedicated layers for hardware integration, controls & navigation, motion planning, and perception worked on developing software for Autonomous Underwater Vehicle (AUV) Show less

• Integrated Robocomp, a robotic framework, with a 3D robotic simulator, Gazebo, using zeroc-ice as a communication middleware.• Used Gazebo plugins for robotics interfaces, corresponding to different sensors and actuators, to communicate with the Gazebo simulator.• The integration is expected to allow developers more options from the framework and provide a better simulation with a more realistic physics engine.

• Explored the possibilities of developing a lightweight traffic light detection model based on Deep Learning, using various model compression techniques.• Explored various techniques of reducing the size of neural networks without any significant decrease in accuracy.• Implemented an RFCN network using TensorFlow object detection APIs.

• Worked as a front end developer to develop new features and improve UI/UX of a scalable web application.• Used the latest technology stacks like TypeScript in Angular 6 as well as HTML and SCSS for styling.

• Promoted Robotics in the campus community by organizing workshops and lectures• Assisted the Coordinators in organizing competitions in Major technical events

• Promoted Fine Arts in the campus community by organizing various competitions and exhibitions• Assisted in development of various art pieces around the campus