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.

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.