India
Limitless software engineering in an AI driven world
Bare metal on GCP - https://cloud.google.com/bare-metal/docs
- Joined Amazon after getting a return offer with the Digital Publishing team. - Worked on resolving several business impacting customer-facing issues as part of the on-call rotation. Also implemented fixes to prevent same issues from occurring. - Worked on a script which validates data between two data stores by listening to a queue(SQS) which has an average traffic of around 3 million messages an hour. - Working on a product creation tool that can be used by consumers to create digital products they want to sell at Amazon. - Tech stack: Java, Spring, Guice, TypeScript, React, DynamoDB, SQS, and SNS.
- Worked with the Digital Publishing team making customer obsession the forefront of any development. The team is responsible for handling metadata updates to digital products(Ebooks, Apps for example) at Amazon. - Improved and modified multiple services which have several million requests per hour of traffic. The changes immediately impact millions of customers of Amazon. - Designed and implemented an API that covers the entire set of services of the team to provide status for a metadata contribution made by a customer. The API provides real-time status for all incoming requests to the main ingestion service of the team. - Involved with designing complex architectures of services that handle several million requests an hour. - Tech stack: Java, AWS, Spring, Guice, and internal frameworks.
-Research on the usage of Generative Adversarial Networks (GAN) for thermal to visible face generation. The generated face can then be used to do face recognition. -Implemented a Pix2Pix GAN using TensorFlow and Keras which was trained on the TUFTS thermal-visible dataset for 150 epochs. -The generator is based on UNet with skip connections and the discriminator is based on PatchGAN. -Introduced batch normalization to counter the problem of vanishing gradients and mode collapse associated with GAN’s. -Analysed the quality of images generated by the GAN by using state of the art metrics such as BRISQUE and NIQE. - Studied and analysed over 20 papers from top Journals and Conferences to select the best method to tackle the problem of heterogeneous thermal to visible face recognition.
- Worked primarily on the Needsmart application back-end which was built using Ruby on Rails, PosgreSQL, Redis and Docker. - Developed a robust notification system which incorporated Firebase Cloud Messaging and could notify up-to 1000 users with a single function call. - Decreased the size of response by 50% for listing items for an outlet by writing a grouping feature which groups items based on name and brand. - Developed a global search feature for items, which had an average response time of 60ms for a search query. - Wrote unit tests and integration tests for models and controllers using Rspec that covered 91% of the code base.
-Worked as a Campus Ambassador and Marketing specialist for TedXChennai 2019 edition. -Worked in multiple teams to organize the event which had close to 1000 people in attendance from 20+ countries