Jan 2024 – May 2025 · 1 yr 4 mos · Atlanta, Georgia, United States · On-site

• Working at SAIL Lab under Prof. Alexey Tumanov, exploring various strategies to speedup LLM inference on large scale.

Jul 2022 – Jul 2023 · 1 yr · Hyderabad, Telangana, India · Hybrid

• Developed features as part of Microsoft Purview's Data Lifecycle Management Team to help customers comply with data lifecycle regulations.
• Key accomplishments and contributions at Microsoft:
• Accomplished automation of file preservation workflow for files shared as cloud attachments benefitting thousands of tenants and safeguarding files on the order of 100k and more.
• Improved infrastructure efficiency and performance by enhancing SharePoint C++ backend architecture to reduce SQL calls by 70% in production when deleting files under DLM preservation holds.
• Optimized certain operations such as batch file deletions on SharePoint UI through caching mechanisms reducing SQL calls by 7% in production.
• Resolved critical bugs, performed root cause analysis of critical escalations affecting users worldwide and made relevant code changes, and participate in PR reviews to enhance E2E workflow.
• Participated in various events like:
• Led various hackathon projects like NFT Marketplace involving development of Smart Contracts in Solidity and developed domain-specific Retrieval Augmented Generation pipeline to help OCEs quickly find solutions to customer escalations.
• Conducted knowledge sharing sessions on Blockchain, Machine Learning and Deep Neural Networks.

Jun 2021 – Aug 2021 · 2 mos · Bengaluru, Karnataka, India · Remote

• Developed models based on LSTMs and transformers in PyTorch to paraphrase text given as input.
• Conducted survey of 20+ research papers and open source repositories to find relevant state-of-the-art models and datasets for paraphrase generation.
• Improved these models by modifying architecture, training and fine-tuning strategies to perform well on Quora Question Pairs (QQP) dataset.
• New models outperformed state-of-the-art techniques by 4-26% across different metrics like BLEU, ROUGE, METEOR and CIDEr.