Jun 2022 – May 2025 · 2 yrs 11 mos

• Google Shopping: Large Scale LLM Inference
• Developed algorithms and LLMs for extracting product attributes from millions of merchant webpages daily.
• Architected LLM extraction pipelines and domain-specific XPaths, enhancing coverage and precision.
• Reduced human review time from hours to seconds, significantly improving operational efficiency.
• Multimodal Vision: Developed a smart selector API ensembling instance, semantic, and panoptic segmentation models for accurate background replacement in Ads.
• Generative AI for Google Support [March 2022 - Oct 2023]:
• Improved support agent efficiency by optimizing LLM-based Smart Reply and Smart Compose models, achieving +23% correct case resolution.
• Built training data pipelines for RLHF and fine-tuning, capturing discrepancies between human agents and model responses.
• Designed PII redaction algorithms for generative models, blocking 10x more sensitive data (credit cards/names) than previous baselines.

Jan 2020 – Jun 2022 · 2 yrs 5 mos

• MUM & Search Ranking:
• Integrated Multitask Unified Model (MUM) for context-aware query transliteration, reducing Word Error Rate by 50% and increasing code-switching accuracy by 12%.
• Deployed changes impacting 6% of all search results in India by integrating transliteration into ranking.
• Knowledge Graph & Infrastructure:
• Mined entity names from user signals, adding 25M+ high-precision names (>95%) to the Knowledge Graph.
• Led development of an ensemble transliteration library for 9 Indian languages used by Suggest and Search.
• Built script-agnostic Unicode LSTM based similarity models and co-authored design docs for unigram-to-n-gram migration.

Jul 2018 – Jan 2020 · 1 yr 6 mos

• Agent Console Platform: Architected microservices and a platform serving 20M+ customers for critical lifecycle events (Lost Sim, Disconnect, Change Number).
• Designed and integrated a B2B quote generation platform using React, Jest, Kotlin, Spring, and PostgreSQL.
• Integration: Led integration with Salesforce using Apex, Java, Spring, and Vlocity

May 2017 – Jul 2017 · 2 mos · Noida Area, India

• Created a Deep Learning Classifier to identify and segment video streams into 15 sports and advertisements categories.
• Developed an app for scene change, camera panning and zoom in or out by analysing RGB histograms of video frames using open-cv and optical flow estimation techniques.

May 2016 – Jul 2016 · 2 mos · Gurgaon, India