Runtime/Robotics team.

• Improved ad quality and coverage for low-resource languages (like Hindi, Polish etc.) having low ad coverage but high query volume.

• Collaborated with the Google Research team to finetune query translation models designed specifically to increase ad revenue by A/B testing against production models.

• Worked on the ad-serving infrastructure to process search query signals, retrieve more ads and filter irrelevant ads while ensuring minimal increase in end-to-end ad-serving infra latency.

• Developed a semi-automated machine learning pipeline to extract crucial transaction information from PDF files.

• Built a clustering based tool as a part of the data-generation pipeline to soft label lines in PDFs using self-organizing maps.

• Reduced PDF parsing time from 3 days to approximately 2 hours.