Jan 2024 – Present · 2 yrs 5 mos

• 1. Designed and deployed ML- and DL-based recommendation systems to improve product discovery, with a focus on solving the cold start problem for new products.
• 2. Developed explore strategies using reinforcement learning and representation learning techniques to enhance user engagement through relevant and diverse content exposure.
• 3. Engineered multi-objective ranking models that optimize for CTR, conversion, and content diversity using deep neural networks, product embeddings, and sequence-aware models.
• 4. Built real-time pipelines and inference systems enabling low-latency recommendations and personalized feeds across the Meesho app ecosystem.
• Key Projects:
• 1. FY Feed CTR-Based Ranker:
• Developed a deep learning-based click-through rate prediction model to optimize the FY feed ranking. Utilized user-product interaction features, session embeddings, and attention-based architectures to drive engagement and conversions.
• 2. Search Real-Time Sequence-Based Ranker:
• Built a sequence-aware ranking model for Meesho search, leveraging transformer architectures and user action sequences. Enabled real-time personalization by integrating historical session data with product representations.

Jan 2022 – Dec 2023 · 1 yr 11 mos

• 1. We launched Explore ML ranker for new onboarded products. This construct resulted 2x jump in orders and views contribution and gave better visibility of new products on FY-feed and PDP.
• (LTR, LightGBM, Pyspark, Python, AWS)
• 2. Made a user personalized feed on Homepage for explore products using candidate generation. This model improved explore GMV and Vc.
• (Real time Candidate generation, Pyspark, ML)
• 3. Catalog ranking offer infra : Platform wide created a single Spark model for sale and deals offers which is being used by data science, tech, and product teams for the application of Ml ranking, Auto boosting and CLP pre-indexing. This model is being used by 7 different charters and easily adaptable for future use cases like offer - prediction, discount depth etc.
• 4. Supply Deals Modelling : Created a dynamic model for 3 hour deals construct which was for 24 hours before. This model can easily track 14 different metrics to track deals performance easily. This model can easily track DOD metrics if something is breaking.
• 5. Sale Inventory supply modelling : created a dynamic inventory dashboard for Sale inventory. This dashboard has made a huge impact in terms of predicting inventory on the sale days and helped suppliers to get the maximum conversion.

Jul 2021 – Dec 2021 · 5 mos

• 1. Supplier tracking Dashboard: have created a Dynamic dashboard for FOS suppliers.
• 2. Meesho Maha Sale : Have setup supply metrics tracking for MMS and deploy it Presto SQL.