Apr 2024 – Present · 2 yrs 2 mos

• SmartThings Cloud
• ➤ Bixby 3.0 - Developed Home Context and Function Manifest services to generate LLM prompts, providing comprehensive device context (e.g., states, locations, rooms) and supported functions/intents, enabling accurate intent prediction and device execution from the utterance given by user through Bixby.
• Samsung Ads - Identity Graph
• ➤ ML Device Graph - To address high TV household churn and improve device stability, I led the integration of an ML-powered device graph into our existing pipeline. By leveraging IP colocation algorithms and household clustering, accurately predicted household assignments for devices, boosting TV device stability within households by 3%. Additionally, optimized both infrastructure and code, reducing model training costs by over 50%, resulting in significant efficiency gains.
• ➤ Profile store - Designed and built a unified customer database hosted on Snowflake, integrating device attributes from multiple producers and external vendors.This solution serves as a single source of truth for customer insights, through access to both real-time and historical version-controlled snapshots.
• ➤Developed a Golang-based Profile Service to streamline the onboarding of new data producers and consumers to the Profile Store. The service registers schemas in a Profile Metadata Store on AWS RDS, facilitating a scalable and efficient integration process.
• ➤Designed and implemented key components of the Household Graph for Samsung Ads, enabling precise device-to-household mapping and improving targeting accuracy. By assigning Household IDs to devices, this solution delivered a 5% increase in stability over IP-based methods, leading to enhanced household-level targeting.
• ➤Engineered the Calendar Effectivity Store to track daily device movements within households, providing actionable insights for accurate reporting. This system captured device footprints, uncovering patterns in device mobility.

Apr 2022 – Mar 2024 · 1 yr 11 mos

• Samsung Ads - Identity Graph
• ➤ Combiner Optimization - As Legacy clusters struggled to handle the load, a Spark Streaming job (combiner) was unable to process all the messages produced to Kafka topics in the LGA and SJC datacenters. As a result, the identity graph was losing critical data, shrinking the device universe and impacting the reach and stability of TV devices. To address this, developed a batch job that offloaded specific topics from the Kafka streaming job, leading to a 58% increase in input consumption rate and a significant reduction in data loss.
• ➤ Compliance Manager - To ensure compliance with stringent data protection laws like CCPA and GDPR, built a Compliance Manager to efficiently handle the deletion of data for opted-out devices on a daily basis. Redesigned existing Spark jobs to construct a rolling 30-day data universe, processing only daily incremental data. This approach eliminated the need for daily snapshots of device data, reducing AWS S3 storage requirements by 80% while maintaining compliance with privacy regulations.
• ➤ Data Observability Framework - Introduced multiple validation rules and setting thresholds for key metrics such as data freshness, completeness, and distribution. This proactive monitoring approach reduced the incidence of DAG (Directed Acyclic Graph) failures and unhandled data anomalies by 99%, significantly enhancing data quality, reliability, and the overall stability of data pipelines.

Jan 2021 – Mar 2022 · 1 yr 2 mos

• Samsung Ads - Identity Graph
• ➤3rd party Device Graph Integrations - Spearheaded the integration of the RoqAd Device Graph with the existing Device Graph, driving a 30% increase in audience reach across the EU region. This integration enhanced targeting capabilities, improved personalized content delivery, and significantly expanded the platform's ability to engage users, resulting in higher user acquisition and retention rates in a key market.
• ➤Global Expansion - Redesigned and optimized Apache Airflow DAGs, consolidating multiple workflows into a streamlined master-slave design. This improvement resulted in a 70% reduction in maintenance and expansion effort, significantly increasing operational efficiency. Additionally, enabled rapid global expansion of the identity graph, making it configurable per region and simplifying the process, leading to faster deployment and scalability.

May 2019 – Jul 2019 · 2 mos · Bengaluru, Karnataka, India

• Worked with Samsung Pay mini payment team and analyzed the payment data of paytm , mobikwik and freecharge collected through Samsung pay mini app.