Reya Sadhu
Machine Learning Engineer/Data Scientist with 2+ years of experience
Experience
June 2024 - September 2024
Machine Learning Engineer Intern @ TikTok
Seattle, Washington, USA
ETA Models for TikTok Shop E-commerce Logistics
• Implemented new features through statistical analysis of distributions, achieving a 30% reduction in prediction range and simultaneously lowering breach rates from 6% to 4.5%
• Implemented two phase multi label classification model in Pytorch to predict delivery breaches: initial phase reduces feature space and do a point estimation, followed by a multi-head architecture to estimate errors around the estimate.
• Developed a hierarchical multi label classification model in tensorflow, incorporating local and global loss.
April 2024 - June 2024
Graduate Student Researcher @ UCSD
California, USA
Language Model for Mutation Prediction
• Designed a Decoder-only Transformer model in Pytorch to predict the next mutations in SARS-CoV-2 genomic sequences, leveraging masked self-attention.
• Conducted research on various weight initialization techniques (Xavier, Kaiming, normal) and activation functions to mitigate vanishing and exploding gradient issues in deep neural networks.
June 2022 - September 2022
Data Scientist @ Fractal Analytics
Bangalore, India
• Market Mix Modeling: Developed of a Hierarchical Bayesian Regression model to decode contribution of KPIs, ROI of sales drivers and provide insights for improving Organic Search Ranks on Amazon e-commerce.
• Generative AI: Developed a POC on the application of Computer Vision in cosmetic surgeries using Generative Adversarial Network (GAN) with landmark detection for face reenactment and post-operative face estimation.
• Time Series Forecasting: Developed time-series forecasting models utilizing ARIMA & VAR and integrated with a postgreSQL database to predict different components of P&L, which resulted in tax reduction.
• Explainable AI: Developed, optimized, and deployed an Azure-based ML pipeline on surrogate models (SHAP/LIME) using Synapse- ML on Python and PySpark scripts to obtain an interpretable and visual explanation of black box models.
• Clinical Trail Pipeline: Developed a POC of a clinical trial pipeline utilizing scispaCy to assess patient attributes for optimal patient selection and cluster formation. Created synthetic clinical data and used CNN for medical image analysis.
Projects
Personalized E-commerce Recommendation
A retrieval based recommendation system with two tower architecture.
Review Rating prediction
Modified neural matrix factorization model for Amazon Products combining item metadata, review texts, and user-item interactions.
Multitask Learning BERT
A multitask model developed by fine-tuning pre-trained BERT on three downstream tasks.
CUDA convolution Filter
Custom CUDA kernel for Image Filtering, integrated with PyTorch to incorporate GPU accelerated custom operation in deep learning workflow.
Image Segmentation
Pattern classification model from scratch using Parametric Estimation and EM Algorithm.
Optimization Algorithm Comparison
Detailed report on Gradient Descent Optimization Algorithms with supporting Python script.
Repeat Purchase Prediction
An XGBoost Classifier based model to estimate the likelihood of a customer making a repeat purchase.
Scene Classification
Ensemble model by leveraging transfer learning with CNNs for image classification.
Education
University of California San Diego
August 2023 - March 2025
M.S. in ECE: Machine Learning & Data Science, GPA: 3.5/4
• Relevant Coursework: Natural Language Processing, Visual Learning, Computer Vision, GPU Programming, Statistical Learning, Parameter Estimation.
• Proficient in statistical analysis, hypothesis testing, and applying machine learning algorithms for predictive modeling.
• Developed skills in visual data processing and parallel programming for efficient large-scale data processing and ML tasks.
Indian Institute of Technology Bhubaneswar
July 1017 - May 2021
B.Tech.(Honours) in Electrical Engineering, GPA: 3.91/4
• I have undertaken coursework in CS including DS and Algorithms, and Software Engineering and electives in Machine Learning, Information Theory and Image Processing enabling me proficient in C/C++, Python, DBMS, OS.
• I have done several projects in classification and regression models, deep learning, and image classification.
• I achieved the second-highest CGPA among a cohort of 60 students majoring in Electrical Engineering.