Dec 2024 – Sep 2025 · 9 mos · Bengaluru · On-site

• As part of the ShopX team, I grew into leading search relevance initiatives after driving personalization and recommendation systems as an IC, partnering with product and engineering to deliver solutions that boost conversion and profitability. Key contributions include:
• 𝗦𝗲𝗮𝗿𝗰𝗵 𝗥𝗲𝗹𝗲𝘃𝗮𝗻𝗰𝗲
• ⊡ Query Correction Layer (QCL), Intent Prediction, Vector Search, Query Segmentation and Experience.
• 𝗣𝗲𝗿𝘀𝗼𝗻𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻
• ⊡ Designed and deployed 𝗚𝗢𝗗 (𝗚𝗼𝗼𝗱𝘀 𝗼𝗳 𝗗𝗲𝘀𝗶𝗿𝗲) products, a new personalization primitive using LightFM, powering user-tailored item suggestions on the “𝗔𝗱𝗱 𝗠𝗼𝗿𝗲 𝗜𝘁𝗲𝗺𝘀” module on the 𝗖𝗮𝗿𝘁 𝗣𝗮𝗴𝗲.
• ⊡ Built and deployed new sub-category activations using LightFM to personalize user–subcategory affinity and surface relevant deal products that users haven't purchased before.
• ⊡ Developed predictive carts for active users using GPT-4 Mini, and prebuilt carts for first-time buyers using FP-Growth.
• ⊡ Developed a POC using a two-tower architecture to improve personalization.
• 𝗥𝗲𝗰𝗼𝗺𝗺𝗲𝗻𝗱𝗮𝘁𝗶𝗼𝗻𝘀
• ⊡ Developed cross-sell reco systems for the 𝗦𝗲𝗮𝗿𝗰𝗵 𝗥𝗲𝘀𝘂𝗹𝘁𝘀 𝗣𝗮𝗴𝗲 (𝗦𝗥𝗣), including: Query–Query and Query–Item recos using FP-Growth, as well as Item–Item recos using Prod2Vec, later enhanced with GNNs for new categories.
• ⊡ Developed similar and orthogonal product reco systems on the 𝗣𝗿𝗼𝗱𝘂𝗰𝘁 𝗗𝗶𝘀𝗽𝗹𝗮𝘆 𝗣𝗮𝗴𝗲 (𝗣𝗗𝗣) using CLIP for similarity and GPT-4o for orthogonal discovery.
• ⊡ Collaborated on the “𝗦𝘄𝗮𝗽” feature (Swap and save, Swap for better value, Swap and go premium, Swap and go organic), driving supplementary recos using NV Embed V2 and heuristics, and the “𝗔𝗱𝗱” feature (Add and save), driving complementary recos on the 𝗖𝗮𝗿𝘁 𝗣𝗮𝗴𝗲 using BERT4Rec and heuristics.
• 𝗢𝘁𝗵𝗲𝗿 𝗜𝗻𝗶𝘁𝗶𝗮𝘁𝗶𝘃𝗲𝘀
• ⊡ Collaborated on the browse ranking system and attribute-based filtering (e.g., Premium, Best Sellers, Trending, Organic, Combos) to improve discovery and conversion.

May 2024 – Dec 2024 · 7 mos

• Led the 𝗥𝗘𝗖𝗞𝗢𝗡 framework and its applications, 𝗖𝗬𝗚𝗡𝗨𝗦 (𝗨𝘀𝗲𝗿 𝗣𝗲𝗿𝘀𝗼𝗻𝗮𝗹𝗶𝘇𝗲𝗱 𝗘-𝗠𝗮𝗶𝗹 𝗡𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿 𝗮𝗻𝗱 𝗢𝗻𝗦𝗶𝘁𝗲 𝗥𝗲𝗰𝗼𝗺𝗺𝗲𝗻𝗱𝗮𝘁𝗶𝗼𝗻𝘀) and 𝗖𝗟𝗢𝗦𝗥 (𝗖𝗼𝗻𝘁𝗲𝗻𝘁 𝗟𝗲𝘃𝗲𝗹 𝗢𝗻𝗦𝗶𝘁𝗲 𝗥𝗲𝗰𝗼𝗺𝗺𝗲𝗻𝗱𝗮𝘁𝗶𝗼𝗻𝘀). Managing release cycles and driving feature enhancements for continuous optimization and business impact.
• ⊡ Led cross-functional collaboration with engineering and product teams to scale and expand 𝗖𝗬𝗚𝗡𝗨𝗦 and 𝗖𝗟𝗢𝗦𝗥 projects, supporting multiple channels and enhancing user engagement.
• ⊡ Expanded 𝗖𝗬𝗚𝗡𝗨𝗦 with A/B testing and advanced candidate generation and reranking strategies to improve content relevance.
• ⊡ Enhanced 𝗖𝗟𝗢𝗦𝗥 by incorporating additional entities and temporal features, and extending to Conde Nast Traveler.
• 𝗪𝗮𝗻𝗱𝗲𝗿𝗹𝘂𝘀𝘁 - 𝗧𝗿𝗮𝘃𝗲𝗹 𝗜𝘁𝗶𝗻𝗲𝗿𝗮𝗿𝘆 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗼𝗿 (𝗣𝗢𝗖)
• ⊡ Developed a POC utilizing Generative AI to craft personalized travel itineraries based on individual preferences such as destination, budget, and interests, with a focus on inclusivity and accessibility. The generator provides daily schedules, activity descriptions, relevant links, and quick actions, targeting Gen Z travellers who seek unique, budget-friendly, and social-media-ready experiences.

Oct 2021 – Apr 2024 · 2 yrs 6 mos

• Led the 𝗥𝗘𝗖𝗞𝗢𝗡 framework and its application 𝗖𝗬𝗚𝗡𝗨𝗦 (𝗨𝘀𝗲𝗿 𝗣𝗲𝗿𝘀𝗼𝗻𝗮𝗹𝗶𝘇𝗲𝗱 𝗘-𝗠𝗮𝗶𝗹 𝗡𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿 𝗥𝗲𝗰𝗼𝗺𝗺𝗲𝗻𝗱𝗮𝘁𝗶𝗼𝗻𝘀), by strategizing feature enhancements for continuous improvement.
• ⊡ Developed efficient hyperparameter tuning using Bayesian search optimization methods.
• ⊡ Developed performance dashboards, designed A/B tests, and advanced reranking strategies to boost content relevance and engagement.
• ⊡ Contributed to offline testing strategies (Backtesting, Counterfactual Evaluation).
• 𝗖𝗟𝗢𝗦𝗥 (𝗖𝗼𝗻𝘁𝗲𝗻𝘁 𝗟𝗲𝘃𝗲𝗹 𝗢𝗻𝗦𝗶𝘁𝗲 𝗥𝗲𝗰𝗼𝗺𝗺𝗲𝗻𝗱𝗮𝘁𝗶𝗼𝗻𝘀)
• ⊡ Developed the CLOSR system by leveraging the RECKON framework, significantly improving related article recommendations using a co-occurrence graph. Achieved a notable increase in both offline and online metrics, outperforming the baseline that used RoBERTa embeddings.
• 𝗧𝗵𝗲 𝗡𝗲𝘄 𝗬𝗼𝗿𝗸𝗲𝗿 (𝗧𝗡𝗬) 𝗖𝗮𝗿𝘁𝗼𝗼𝗻 𝗖𝗮𝗽𝘁𝗶𝗼𝗻 𝗖𝗼𝗻𝘁𝗲𝘀𝘁
• ⊡ Developed a Multi-Armed Bandit (MAB) model to optimize caption selection for The New Yorker Cartoon Caption Contest, enhancing reader engagement by dynamically presenting diverse, bias-free caption submissions. Focused on improving user interaction and identifying the most humorous captions for finalist consideration.
• 𝗙𝗨𝗦𝗜𝗢𝗡 - 𝗜𝗺𝗮𝗴𝗲 𝗦𝗶𝗺𝗶𝗹𝗮𝗿𝗶𝘁𝘆 𝗦𝗲𝗮𝗿𝗰𝗵 (𝗣𝗢𝗖)
• ⊡ Created a sophisticated visual image and object search solution utilizing ResNet-50 for image embeddings and YOLOv5 for object detection. This system enables users to search for similar images and objects, with models served via Databricks model endpoints. Delivered improved accuracy and usability for complex image searches.

Jul 2020 – Sep 2021 · 1 yr 2 mos

• 𝗚𝗿𝗮𝗽𝗵 𝗘𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝘀 𝗶𝗻 𝗣𝗲𝗿𝘀𝗼𝗻𝗮𝗹𝗶𝘇𝗲𝗱 𝗥𝗲𝗰𝗼𝗺𝗺𝗲𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗦𝘆𝘀𝘁𝗲𝗺𝘀 (𝗣𝗢𝗖)
• ⊡ Conducted an in-depth exploration of Knowledge Graph embeddings in Personalized Recommendation Systems, testing traditional methods like TransE, TransR, Node2Vec, and various Graph Neural Network (GNN) models. Achieved significant performance improvements compared to Doc2Vec embeddings, demonstrating enhanced accuracy in recommendations.
• 𝗥𝗘𝗖𝗞𝗢𝗡 (𝗥𝗘𝗖𝗼𝗺𝗺𝗲𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗦𝘆𝘀𝘁𝗲𝗺𝘀 𝘂𝘀𝗶𝗻𝗴 𝗞𝗻𝗢𝘄𝗹𝗲𝗱𝗴𝗲 𝗡𝗲𝘁𝘄𝗼𝗿𝗸𝘀) 𝗳𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸.
• ⊡ Led the design and development of the RECKON framework, employing an encoder-decoder architecture to encode entities—such as articles, authors, and users—into graph embeddings. The system utilizes GraphSAGE as the encoder and MLP/DotProduct as the decoder to provide precise recommendations.
• 𝗔/𝗕 𝗧𝗲𝘀𝘁𝗶𝗻𝗴 𝗦𝘂𝗰𝗰𝗲𝘀𝘀 𝘄𝗶𝘁𝗵 𝗥𝗘𝗖𝗞𝗢𝗡 𝗶𝗻 𝗖𝗬𝗚𝗡𝗨𝗦
• ⊡ Executed A/B testing of RECKON-based recommendations within the CYGNUS Personalized Email Newsletters, resulting in a 15% increase in CTR over the existing Two-Tower recommendations model, showcasing its effectiveness in driving user engagement.
• 𝗦𝗣𝗜𝗥𝗘 - 𝗦𝗲𝗴𝗺𝗲𝗻𝘁𝗲𝗱 𝗔𝗱-𝗧𝗮𝗿𝗴𝗲𝘁𝗶𝗻𝗴 (𝗣𝗢𝗖)
• ⊡ Improved models by utilizing shallow encoded knowledge graph embeddings to enhance existing ad-targeting models. Additionally, I created Databricks SQL dashboards to monitor campaign performance and analyze user segment metrics, further refining targeted advertising strategies.

Jan 2020 – Jun 2020 · 5 mos

• 𝗦𝗣𝗜𝗥𝗘 (𝗦𝗲𝗴𝗺𝗲𝗻𝘁𝗲𝗱 𝗔𝗱-𝗧𝗮𝗿𝗴𝗲𝘁𝗶𝗻𝗴)
• ⊡ Improved the models through video-related feature integration, actively contributing throughout the data science lifecycle - encompassing data collection, storage, processing, feature engineering and modelling.
• ⊡ Created databricks dashboards to assess third-party data quality and compare model backtesting results, streamlining query execution times and automating processes for new incoming trait ids.
• 𝗦𝘁𝗮𝗻𝗱𝗮𝗿𝗱𝗶𝘇𝗮𝘁𝗶𝗼𝗻 𝗼𝗳 𝗔𝗱𝘃𝗲𝗿𝘁𝗶𝘀𝗲𝗿 𝗡𝗮𝗺𝗲𝘀
• ⊡ Developed a POC for standardizing (canonicalization) advertiser names from diverse sources, tackling challenges of large dataset size and lack of unique labels. Leveraged unsupervised learning techniques including sparse-dot-topn and FastText for character-level embeddings.

May 2019 – Jul 2019 · 2 mos · Chennai Area, India · On-site

• Worked as a summer intern in the Data Engineering team, at Condé Nast Chennai. I have worked on the following:
• ⊡ Worked with Facebook GraphAPIs to collect data from Facebook pages using Python and Pandas.
• ⊡ Setup an endpoint using ExpressJS to receive webhook events (real-time updates in Facebook page feed) in the form of JSON, parsed those objects and stored them in a CSV file using data frames provided by dataframe-js and deployed this application in AWS EC2.
• ⊡ Worked on an end-to-end case study of sentiment analysis on user comments on the Vogue Facebook page: To predict whether a comment is positive, negative or neutral. (a multi-class classification problem)

Mar 2016 – Feb 2017 · 11 mos · Tekkali

• ⊡ I worked as a head of this club in my final year of graduation and my responsibilities include training juniors on web-technologies and guide them in building web applications.
• ⊡ I have also conducted a few workshops on Web Development in different engineering colleges.

Mar 2015 – Feb 2016 · 11 mos · Tekkali

• ⊡ WebApps is a club from the student activity centre (Leadership) in which we learn and build applications related to web.
• ⊡ My responsibilities include maintaining the SAC websites and designing the websites for fests and events along with the team.