Work Experience

Professional Experience



Data Science Intern, Growth Data Science Team

May 2024 - Aug 2024

Moloco

  • Increased win rates by 5% and actions by 30%, improving campaign performance for AMR consumer advertisers by adjusting discount rates dynamically for high-value users. Boosted CPA predictability by 20%through statistical research on creative diversity, funnel category, and budget mode, leading to actionable A/B test recommendations.
  • Achieved a 3-4% increase in ROAS for Activision by correlating pLTV with purchase events, leading to better revenue performance in D3 pLTV campaigns.
  • Developed DMA-based strategies for Bumble that reduced CPA by 15%, identifying optimal spend levels and high-performing DMA markets, enhancing profitability for UA campaigns.



Data Science Capstone

Jan 2024 - May 2024

Sabre Labs

  • Lead the development of a robust LLM-based solution, leveraging LlaMA 2 to achieve a unified address structures, by processing 1.35 million records of lodging properties from diverse aggregators. Acheived an initial accuracy score of 82.7%
  • Utilized a BERT-based model to generate contextual embeddings from categorical data in over 6 million shop requests to Sabre IntelliSell, improving accuracy and performance for cache rate prediction.
  • Utilized generated contextual embeddings for dynamic price prediction as a downstream task.
  • Leveraged the HPC capabilities of the TACC-Lonestar 6 supercomputer for parallel computing and optimization.



Data Science Intern

April 2021 - July 2021

Twimbit

  • Data Analysis: Installed and managed a holistic data pipeline (Algolia, Heap, Matomo, Segment) for tracking website user interactions to facilitate data-driven decisions.
  • Leveraged A/B test insights and ad-hoc analysis to reduce product friction and boost daily user numbers by 5%.
  • Machine Learning: Parsed raw HTML data from 700+ webpages on the product website using Beautiful Soup to train a Decision Tree model for automated classification of records into distinct categories.
  • Proposed and implemented a unique metric correlating read time to page depth scrolled, improving page readability and user retention by 20%.
  • Utilized text processing and topic modeling using gensim and spacy-transformers, leading to a 64% improvement in search query response time.

Projects

Reinforcement Learning for multi-node pricing and inventory management.

  • Implemented and compared the performance of RL methods (A2C, DQN, PPO) against conventional Mixed IP optimized pricing and (S, s) re-order policies for single/multi echelon environments with stochastic demand.
  • Acheived an impressive 25% increase in profit with dynamic pricing, and 1.09 times increase in profit compared to traditional (S, s) order policy.

    • Multi-Modal Content Generation and Alignment with Efficient Optimization.

  • Seamlessly merged a Stable Diffusion based text to image model with AlignProp (a backpropagtion based approach to align diffusion models with downstream reward functions) and DreamGaussian (a 3D Gaussian Splatting model with mesh extraction and texture refinement).
  • Demonstrated the superiority of DreamGaussian in generating high-quality textured meshes, achieving a remarkable 10-fold acceleration compared to existing methods, with the ability to produce content from a single-view image in just 2 minutes.

    • A graph-based big data optimization approach using Hidden Markov Model and Constraint Satisfaction Problem.

  • This project was undertaken as a requirement for the PhD level Graduate Course 'Statistical Modeling 1' at UT Austin, supervised by Dr. Abhra Sarkar, Assistant Professor - Department of Statistics and Data Sciences, UT Austin
  • Developed a framweork to incorporate a Constraint Satisfaction Problem in reducing the state space of the Hidden Markov model with use in financial time series analysis. Used this model to predict closing prices of the Dow Jones Industrial Average
  • Induced constraints into the model by analyzing Twitter tweet sentiments over the period Jan 20, 2017, to Jan 20, 2020. This varied approach from a conventional model reduced the Mean Average Percentage Error (MAPE) by 0.59%, increased accuracy to 90%, and reduced the computational time by 0.02s
    •

    Deep Learning integrated with modeling a medical waste gasification-power production plant

  • This work was a part of my Undergraduate Thesis, supervised by Dr. Dinesh Shankar Reddy (Associate Professor, NIT Andhra Pradesh)
  • Conducted literature surveys on the sources, types, effects, of medical waste generated, and the various properties associated with characterizing medical waste
  • Developed a Deep Learning model using PyTorch, to simulate thermodynamic equilibrium modeling of a downdraft gasifier for the conversion of medical biomass to energy.
    •