Aug 2023 – May 2025 · 1 yr 9 mos

• Leading RichAds modeling team's efforts in improving the clickability & quality of Search Ad-Extensions & Dynamic Search Ads (DSA) across EMEA, APAC & LATAM markets.
• Introduced click-prediction (CP) models for extensions & DSA headlines utilising historical signals, contributing 3-6% revenue increase across tiers.
• Developed a global feature-store, upgrading on the region-specific design, extending ranking service from 5 to 100+ markets.
• Incorporated semantic query-context signals into CP model, resulting in a +3% Delta AUC on impressed ads.
• Curated an offline selection approach for autogenerated extensions using historical query-context features. Exploited marginalised scores from a semantic CP model while allowing for random exploration. Scaled & globalised this pipeline, enabling daily ranking of ~10B items.
• Addressed a combinatorial variant ranking problem by formulating a theoretical approach, conducting large-scale hypothesis testing, defining features, and using DCNv2 model. This resulted in an offline +4% Delta AUC on impressed ads.
• Designed an E2E personalisation paradigm leveraging long-term and real-time user-interest signals to provide a personalised ranking scheme, making the items more relevant and diverse.
• Technology & Tools: MapReduce (Scope), PyTorch, Keras, Huggingface, ONNX, Pandas, Matplotlib, SciPy, NumPy, SkLearn, LangChain, PySpark, Jupyter, Docker, Kubernetes, Tensorboard, WandB, Azure Data Factory, Azure Data Lake Storage, Azure Data Lake Analytics, Distributed FS (Cosmos), Kafka, BLAS, LaPack, Intel MKL, GDB, JDB, Valgrind, Perf, RESTful APIs, OAuth, Git, Conda, Pip

May 2021 – Aug 2023 · 2 yrs 3 mos

Dec 2018 – May 2021 · 2 yrs 5 mos

• Bengaluru Area, India
• Area: Ad-decorations for bing.com
• My work broadly involved candidate generation (extraction), in particular,
• Algorithm ideation for candidate extraction from various data sources
• Designing/development/deployment/monitoring of extraction pipelines
• A/B testing/analysis of performance/mainstreaming of the newly generated candidates
• On occasions, developing additional back-end plugins for serving in real-time
• Achievements: Quarterly Excellence Award Q4 2019-2020
• Technologies: SCOPE, C#, Python, Azure Data Factory

Dec 2016 – Dec 2018 · 2 yrs

• Area: Oracle Cloud Infrastructure
• Designed & implemented a majority of the Marketplace REST API.
• Employed batch-processing & application-layer caching to reduce the response times of multi-page GET-calls from ~2 mins to ~10 secs.
• Implemented a seamless onboarding workflow of existing SaaS customers to PaaS service offerings within a tenant automation framework.
• Technologies: Java/JEE, Jersey/Jackson, ADF, JPub, PL/SQL, OAuth2.0, WebLogic

Sep 2016 – Dec 2016 · 3 mos

May 2016 – Aug 2016 · 3 mos · London, United Kingdom · Remote

May 2014 – Aug 2014 · 3 mos · Mumbai, Maharashtra, India · Remote

May 2013 – Aug 2013 · 3 mos · Mumbai, Maharashtra, India · Remote

May 2016 – Jul 2016 · 2 mos · London, United Kingdom · On-site

• Worked with Prof Arthur Gretton and Dr. Heiko Strathmann as a research assistant.
• Devised a cache-friendly algorithm for non-parametric two-sample tests involving MMD estimator that showed ~300x speed-up over naive implementation.
• Proposed & implemented a multi-threaded variant that outperformed competing algorithms, built with state-of-the-art solvers, by reducing the runtime tenfold.
• Co-authored a paper where this is utilised in a discriminator for GANs.
• Technologies: C/C++, Cachegrind, Perf, OpenMP, Python

Jul 2014 – Apr 2016 · 1 yr 9 mos · Kalyani Magnum, Bangalore

Jul 2011 – Nov 2013 · 2 yrs 4 mos · Mumbai, Maharashtra, India · On-site

• I worked as a system administrator in Center for Formal Design and Verification of Software Lab.
• I implemented bounded-model-checking using Symbolic Simulation of RTL programs as a part of a library (in C++) developed by our lab for hardware verification via symbolic simulation.