• Enhanced bot responses with semantic caching usinglangchain, pinecone, and GPTCache, reducing averageresponse time from 20 sec to 3 sec in testing• Revamped translation pipelines with a faster and robusthybrid of Google Translate and GPT for a seamlesschatbot experience in any language• Implemented real-time cost and token tracking systemAPI for client for using services like GPT and GCP• Developed an automated pipeline for generating testdatasets, focusing on the bot’s failure cases for continuousimprovement

• Designed and implemented a novel Clustered LayerPersonalized Federated Learning approach formulti-device non-iid distributed data• Boosted performance in comparison to standard FedAvg(≈ 30−40 % ) and Clustered Federated Learning(≈8−10 % ) when tested on non-iid MNIST Dataset• Developed an Automated resource allocation solutionfor distributed learning in the mobile-edge-cloud continuum• The proposed algorithm efficiently handles tasks in order oftheir priority, while minimizing the cost for each task• Worked with Genetic algorithms, ReinforcementLearning, Integer programming and Heuristic Searchapproaches for problem-solving

-Writing articles related to recent developments in machine learning and AI.

-Engineered advanced Deep Learning in Road/Lane detection for self-driving cars-Developed modifications of EDA-Net and Res-Net-Reviewed Literature on Semantic Image Segmentation-Gained 2 months of valuable experience and proficiency in coding DL models in the PyTorch framework

-Worked on Improvising the methods for Unsupervised Semantic Image segmentation over the SOTA approaches-Revamped proposed approaches W-Net to improve the performance on the TGS Salt Identification Challenge dataset-Acquired knowledge of Normalized cuts and hierarchial image segmentation algorithms