Apr 2022 – Jul 2024 · 2 yrs 3 mos

• ❖ Led efforts in evangelization and large scale adoption of generative AI techniques within the org. Worked with the strategy team in evaluating ideas from external startups and helped conduct hackathons/tech sessions to source ideas and build a framework for evaluating relevant ones for adoption. Over the past year we have shipped generative AI lead products across food catalog image/text enrichment, customer review summarisation, conversational analytics for enhanced productivity, AI generated brand videos , & WhatsApp based QnA bot for restaurant partners that has resulted in significant business impact .
• ❖ Drove consensus among stakeholders and lead efforts in building natural language based search . Swiggy’s natural language search enables users to search using conversational and open-ended queries and receive recommendations tailored to their specific needs.
• ❖ Lead the charter on ‘search intelligence’ where I utilized traditional ML as well as generative AI based techniques to mine insights from the large corpus of search data and influence stakeholders to implement product ideas such as ‘swim lanes’ in the app, onboarding of key restaurants and enhancing menu items/categories based on trending topics.
• ❖ Lead the forecasting efforts for estimating daily demand of food/grocery items for Swiggy-Instamart. The ML based forecasting models are tuned to improve availability of sku’s throughout the time of operation of the dark store and minimize wastage. Leaning heavily on my prior experience in building products in forecasting, I guided the team to arrive at a realistic estimate of demand(from sales data), implement anomaly removal techniques, adjusting for price elasticity, SOTA methods for modeling of long tail items that are intermittent and lumpy and implement hierarchical approaches for estimating aggregate demand.

Feb 2019 – Mar 2022 · 3 yrs 1 mo

• ❖ I played a pivotal role in conceptualizing, deployment, and continuous improvement of fraud detection models for handling abusive customer claims , identifying fraudulent delivery agents, & detecting abuse in cash on delivery orders. I was instrumental in the development of a weak supervision-based label generating methodology that works at scale that generates tagged data to power the various fraud models. I also championed adoption of anomaly detection techniques, graph-based community models for detecting co-located fraud, & using multimodal signals for augmenting fraud model performance.
• ❖ Served as the tech lead and help build several foundational capabilities for the ML charter for geospatial intelligence, ranging from extracting POIs, entry/exit gates from maps to defining metrics to estimate and compare the quality of different map networks to using ML to find missing roads in the Open Streets Map network to improving address quality to ML models for estimating point-to-point distances accurately. The outcomes resulted in improved last last mile delivery efficiency, better customer and delivery agent experience.
• ❖ Instrumental in developing speech recognition engine that transcribes calls between various entities in Swiggy’s context and mine insights. Addressed challenges arising from heavy code-mixing, high background noise, lack of tagged data, by implementing novel methods for model training. The resulting model provided good accuracy to power use cases for understanding intent of calls.
• ❖ Key member of the logistics team that developed new capability of cost function-based assignment of orders to delivery agents over the existing greedy approach resulting in major turnaround of delivery system efficiency. Lead efforts for combining order batching and assignment in a single cost function to improve customer experience. Helped build the ML driven ‘promises engine’ that generates predicted arrival time estimates for orders on the platform.

Jul 2018 – Jan 2019 · 6 mos

• ❖ Key member of a global team that worked closely with an airline customer to bring predictive analytics capabilities to address their core demand and fleet optimization challenges. Provided a sales contribution solution that enabled the revenue management team to offer best price for quotes, eventually leading to lift in annual revenues of over 5%. Developed and implemented a ‘Schedule Optimization’ solution for a private aircraft jet client, that tracks the aircraft schedule and based on a matching algorithm, assigns the aircraft to the next best trip option, to minimize ‘deadheads’ or unpaid flights.
• ❖ Part of a global team that helped address the problem of inventory management for a major US based tire distributor across their distribution centers. Helped build demand forecasting solutions that generated category based and sku based demand, using time series methods and its variants. Developed an explanatory model that quantified the impact of weather parameters on demand across various tire categories.

Jul 2016 – Jun 2018 · 1 yr 11 mos

• As an early member of Noodle.ai, contributed significantly to firm building and management involving in tasks like coaching new employees, recruiting key talent, data science process development, training material development, and generating project proposals and demos.

Oct 2013 – Jul 2016 · 2 yrs 9 mos · Bangalore · On-site

• Played a key role in the development and product insertion of acoustic add-on devices that gave LM significant competitive advantage in low noise blade technology.

Dec 2012 – Sep 2013 · 9 mos

• Lead India team on various aspects of aircraft engine noise and worked with a global team to ensure that GE engines meet noise certification standards. Made sizable contributions Un-Ducted Fan (UDF) acoustics & development of models for Low Pressure Turbine (LPT) noise reduction.

Oct 2010 – Nov 2012 · 2 yrs 1 mo

• Lead the acoustics team in understanding noise generated by Low Pressure Turbine (LPT)
• and developed a semi-empirical model for LPT noise prediction that would enable design of
• low noise turbines that would help meet the stringent noise regulations stipulated by the
• FAA.

Apr 2008 – Sep 2010 · 2 yrs 5 mos

• Instrumental in developing a novel narrow band technique for decomposition of engine noise
• spectra which improved noise margin estimations and brought GE process in par with
• industry standards.

Jan 2006 – Mar 2008 · 2 yrs 2 mos