Oct 2023 – Mar 2024 · 5 mos

• Core Labs (Applied AI) @ Google
• I lead the GenAI Retrieval Augmented Generation (RAG) infrastructure area, which includes Vector databases and infrastructure for other semantic retrieval techniques to augment LLM's knowledge for Q&A on private corpora. This infrastructure is being used by several internal and external facing products at Google for solving business-critical problems using RAG.
• Gemini API semantic retrieval is built on top of vector store infrastructure developed by my team: https://ai.google.dev/

Apr 2021 – Oct 2023 · 2 yrs 6 mos

• Generative AI APIs @ Google Labs
• Owned performance eval and improvements for Google PaLM API. Improved TPU inference latencies and e2e API latencies by understanding and experimenting with quantization, batching, model sharding. HBM usage and bandwidth constraints, TPU topologies etc.
• Improved latency of the PaLM API by up to ~10x, making it several times faster than other GenAI API offerings in the industry. The performance work I did was behind multiple GenAI products/features announced at Google I/O 2023 including NotebookLM, Google Makersuite and "Help me write" in Gmail.
• PaLM API: https://developers.generativeai.google/
• NotebookLM: https://blog.google/technology/ai/notebooklm-google-ai/
• Makersuite: https://makersuite.google.com/

May 2019 – Apr 2021 · 1 yr 11 mos

• Tech leading Search Ads Analysis Infrastructure - An extremely fast and reliable infra for A/B experiments analytics.

May 2017 – Apr 2019 · 1 yr 11 mos

• Built a logging and analytics framework for Actions on Google, to enable product/feature teams to compute metrics, platform insights, ranking features and developer analytics for actions on the Google assistant.
• Built efficient, scalable and maintainable infrastructure using Google's bigdata technologies like Flume, Mesa, Bigtable.
• Actively collaborated with 10+ teams and product managers to evolve the infrastructure as per product requirements and help those teams to successfully leverage the infrastructure.
• Key aspects of my work involve batch and streaming data pipelines, logs processing, real time analytics generation, database schema design, SQL query authoring, SQL pipelines, and building datasets that can be sliced and diced.

Oct 2015 – Apr 2017 · 1 yr 6 mos

• Built the backend infrastructure for Ad Balance, a feature that helps publishers show fewer,
• best-performing ads and provide better visitor experience on their sites with minimal drop in earnings.
• Launched new features to the AdSense A/B experiments - automatic A/B experiments and autocomplete A/B experiments.
• Launched new recommendation types for AdSense publishers - Matched content, Account linking with Google Analytics.

Jun 2014 – Jul 2014 · 1 mo · Bangalore

• Worked towards efficient and accurate transductive classification in Hypergraphs using a Heat-Kernel framework.
• Developed a generic mathematical framework for classification in multi-view data represented as multi-layer graphs.
• The framework learns a linear model that combines spectral information from multiple layers using the graph Laplacian.

May 2013 – Jul 2013 · 2 mos · Redmond

• Research Intern, Cloud Systems and Applications Group
• Worked towards an efficient online traffic engineering scheme that exploits information on transfer sizes and deadlines to pack long-running transfers across network paths and time, using a tailored approximate solution to mixed packing and covering problem.
• My work was a part of a paper “Calendaring for wide area networks” that has been published in ACM SIGCOMM 2014.