Title Collection Cohesion• Research: Leveraged LLM (Claude) to detect cohesion amongst an arbitrary group of titles. ImprovedAUROC by 7% and true negative rate by 27%.User-title relevance• Research: Built a Two-Tower Deep Learning Neural Network model to predict user-title relevance.• Tech Lead & Leadership Skills• Cross Functional: Collaborated with multiple teams - Science, Engineering, Merchandisers, Product. • Scaled the team to deliver successful world-wide launch of personalization models.• Roadmap Prioritizations: Prioritized key projects such as real-time inference, unify models for speed, Full CD, new propensities.

Identifying User - Video Taste Affinities• Research: Built a Deep Learning Neural Network with Transfer Learning to predict user tastes.• Research: Built a Two-Tower Deep Learning Neural Network model to predict user sports affinities. • Engineering: Built large-scale (100k requests per second), low latency (10 ms) API.• Engineering: Built automated workflows to process big data (~100TB) using Apache Spark and train ML models.Prime Video Channels Personalization• Research: Built a Logistic Regression model using Transfer Learning to personalize channels.• Engineering: Built large-scale, low latency service and model refresh workflow to personalize channels carousel.OE• Saved $3MM in yearly costs by optimizing bloated library code, EC2 RI allocation, caching KMS keys.

User-title relevance• Research: Added language preference features to Deep Neural Network user-title relevance model. • Research: Extended existing title personalization SVD model to use more latent features.* Engineering: Built and extended large scale data processing workflows to aggregate user behavior, generate training data, train ML models and vend output of ML models online.

Full stack developer working with Java(Jersey), node.js and angular.js.

Full stack developer at the Neurobiology Dept. at UCLA.Developed a MVC app to create graphs of research papers and connect findings of multiple papers.Used node.js, express.js, neo4j (NoSQL) and backbone.js on Amazon Web Services (AWS).

Full stack development of a MVC web application to automate activities during release push.Allow coordination of activities during new release between multiple teams at PayPal.Used node.js, express.js, dust.js, backbone.js, require.js, jQuery, twitter bootstrap and git.

•Developed device servers that represent devices like sensors using Python.•Developed GUI clients (for desktops and smartphones) which interact with the servers and read/write data to devices.

•Deploy ios apps for UCLA Student Media. •Develop scripts in Python, Ruby and UNIX/Linux shell to automate deployment process.

•Worked as developer and debugger at Computational Physical Chemistry Institute.•Developed computationally intensive Monte Carlo simulation programs (Matlab) toinvestigate absorption of Argon on Platinum.•Enhanced performance (x10) by using lookup tables.•Fixed memory leaks (10 MB) in Gromacs, which hindered a wrapper script in Python fromexecuting.•Experience gained in Python, Matlab, C, shell scripting and Valgrind.

Synthesized Ferric Hydroxide and Aluminium Hydroxide Nanoparticles. Used the same to remove Arsenic and Fluoride ions from water. Compared efficiency of nanoparticles to remove ions with precipitates