Worked On several Projects which included the following responsibilities :-I upgraded my team service from CDK (Cloud Development Kit) from version 1 to version 2, enabling the utilization of significant improvements and changes in CDK v2. This involved conducting thorough research, studying documentation, and addressing challenges related to updating dependencies, modifying function calls, and adapting coding conventions. Additionally, I ensured seamless integration by incorporating the updated dependencies into the version set I was working with.In one of the project, my main task was to convert a TensorFlow model and SPM tokenizer to ONNX format and integrate the HuggingFace tokenizer. This conversion enabled the generation of ASIN embeddings, representing unique product IDs in Amazon. Additionally, I enhanced the script to selectively utilize the new model and tokenizer based on the marketplace and introduced a flag for switching between the new and old models. The conversion to ONNX format allowed for more accurate testing in a weblab environment, ensuring consistent tokenization and eliminating known differences in embeddings.I automated the data build process for Q2IBN by creating a script that runs on a cluster after every certain interval of days. The script retrieves queries, retrieves ASINs, computes scores for browse node irrelevance and query-browse node irrelevance, and publishes the resulting builds to S3. This automation ensures timely data availability for other services.

Worked on Developing a highly accurate code-switched text classifier for multilingual languageclassification and Implemented an Online Transliteration System capable of converting words into anytarget language. Generated, trained, and tested data using public resources, utilising a training corpus of30K Hinglish sentences.