Snacks Weekly on Data Science

In this episode, we will explore how Pinterest uses generative AI technology to develop a text-to-SQL feature for their data analytics team. We will examine the general architecture of their two iterations, regarding how the team has enhanced the AI product to support better a wider range of analytics use cases with improved productivity. Based on their published tech blog, with the link provided...

In this episode, we'll discuss what network effects are, how they introduce challenges in the standard A/B testing framework, and how the cluster experimentation method can be leveraged to address these challenges. We will also delve into the technical details of how clusters can be generated, and evaluated, and the associated trade-offs that need to be considered. Based on their published tec...

In this episode, we'll explore how the data science team from Grammarly developed their geo-experimentation to measure marketing effectiveness. We will cover about three components in designing an A/B testing experiment, as well as considerations regarding the opportunistic costs of the experimentation. Based on their published tech blog, with the link provided here for your reference: https:/...

In this episode, we'll explore how the data science team from Shopify leverages Monte Carlo Simulation to develop their sampled success metrics. We'll discuss what is sampled success metrics, the associated trade-offs needed to build them, and how Monte Carlo simulation can be used to inform decisions. Based on their published tech blog, with the link provided here for your reference: http...

In this episode, we'll explore how the data science team from Klaviyo developed a Generative AI Product that enhances experiences and enables efficient customer segmentation. We'll also discuss two key concepts in Generative AI: prompt chaining and few-shot learning. Based on their published tech blog, with the link provided here for your reference: https://klaviyo.tech/building-segments-a...

Description: In this episode, we will talk about the importance of remediating failed workflow jobs to reduce business infrastructure costs. We delve into Netflix's approach, which involves enhancing their existing rule-based error classifier with advanced machine learning models. This allowed for auto-remediation, improving the handling of memory configuration and unclassified errors, ultimat...

In this episode, we will talk about what is technical debt in software engineering and its associated risks. We will also share a set of metrics to measure the status of technical debt and ways to help companies quantify their progress toward better software engineering efforts. Based on their published tech blog, with the link provided here for your reference: https://medium.com/booking-com-devel...

In this episode, we will discuss how the Science team at Amazon designs pricing experiments to improve experimentation power. We will cover concepts like the carryover effect and spillover effect, as well as the solutions the team developed to overcome those challenges. Based on their published tech blog, with the link provided here for your reference: https://www.amazon.science/blog/the-science-o...

In this episode, we will talk about what is position bias in recommendation systems, and how the applied scientist team at Uber tackled this challenge. Based on their published tech blog, with the link provided here for your reference: https://www.uber.com/blog/improving-uber-eats-home-feed-recommendations

In this episode, we will take a look at one analytical approach in decision-making called Analytical Hierarchy Processing (AHP). The New York Times team leverages this AHP process to help with making better collective decisions in their technology choices for privacy. Based on their published tech blog, with the link provided here for your reference: https://open.nytimes.com/collective-decision-ma...

In this episode, we delve into a study conducted by Intuit on measuring the productivity impact of the Gen AI tool on their data analyst team. It demonstrates exciting positive productivity gains, offering an exciting outlook for many in the industry. Based on their published tech blog, with the link provided here for your reference:  ​​ https://medium.com/intuit-engineering/how-intuit-data-analys...

Description: In this episode, we discuss how the data science team at Microsoft designed perturbation analysis to understand the Large Language Model’s performance on commonly seen tasks. Based on their published tech blog, with the link provided here for your reference:  https://medium.com/data-science-at-microsoft/perturbation-analysis-and-llms-how-sensitive-are-llms-to-their-input-91a8407a971f

In this episode, we discuss how DraftKings leverages Monte Carlo simulation to predict tennis game outcomes and generate probabilities to power its online gambling service. Based on their published tech blog, with the link provided here for your reference:  https://medium.com/draftkings-engineering/building-a-tennis-simulation-d6afdaa97d19  

In this episode, we discuss the insights shared by Expedia's Machine Learning Engineering team on how they leverage the two-tower neural network architecture in the candidate generation stage of their recommendation system. Based on their published tech blog, with the link provided here for your reference: https://medium.com/expedia-group-tech/candidate-generation-using-a-two-tower-approach-wi...

In this episode, we discuss how the Engineering team at Walmart created a customized Large Language Model (LLM) agent with Retrieval Augmented Generation (RAG) Technology to improve their agile planning efficiency. Based on their published tech blog, with the link provided here for your reference: https://medium.com/walmartglobaltech/an-autonomous-agent-for-agile-planning-98303e194e08

In this episode, we discuss why and how Intuit leverages automated sanity checks to streamline its machine learning development process. Based on their published tech blog, with the link provided here for your reference: https://medium.com/intuit-engineering/how-to-streamline-ml-model-deployment-automated-sanity-checks-64a23166fdc5

In this episode, we take a look at the learnings from an online grocery tech company regarding their development of machine learning models for scalable demand forecasting. Based on the tech blog from Picnic International, the link is provided here for your reference: https://blog.picnic.nl/running-demand-forecasting-machine-learning-models-at-scale-bd058c9d4aa7

In this episode, we discuss the online-offline discrepancy problem in machine learning models and explore Pinterest's journey to tackle this challenge in their ads ranking system. Based on the tech blog from Pinterest, the link is provided here for your reference: https://medium.com/pinterest-engineering/handling-online-offline-discrepancy-in-pinterest-ads-ranking-system-8fd662da4c2d

In this episode, we explore why and how Instacart’s advertising team unified multiple browse click-through-rate prediction models by leveraging an advanced deep learning architecture. Based on the tech blog from Instacart, the link is provided here for your reference: https://tech.instacart.com/one-model-to-serve-them-all-0eb6bf60b00d

In this episode, we explore how ZipRecruiter, an online employment marketplace, leverages deep learning and Encoder modules to develop their recommendation system, which is designed for better matching job seekers with job listings. Based on the tech blog from ZipRecruiter, the link is provided here for your reference: https://medium.com/ziprecruiter-tech/multimodal-learning-for-employment-marketp...

In this episode, we look at how Oscar Health, a health insurance company, leverages the large language model (LLM), ChatGPT-4, to develop its claim assistant feature. They have built a reasonable assessment system and deployed several strategies to guide GPT-4 in further improvements. Based on the tech blog from Oscar Health, the link is provided here for your reference: https://medium.com/oscar-t...

This episode shares how the applied science team developed an ad simulation system as a more efficient and economical evaluation method for the Uber Eats Ads business. This tool has proven instrumental in guiding and expediting product development and driving growth. Based on the tech blog from Uber, the link is provided here for your reference: https://www.uber.com/blog/unleashing-the-power-of-ad...

This episode shares how the HelloFresh data science team explores the use of generative AI to create new food recipes. Based on the tech blog shared by Data Scientist from HelloFresh, the blog link is: https://medium.com/hellofresh-dev/recipes-and-generative-ai-6d74a107860c

This podcast episode delves into my journey of utilizing unconventional learning methods to fulfill my 2023 New Year's resolution of enhancing my chess skills. The high-level concept and philosophy apply to Learning in general and it aims to inspire others to reflect on their best learning strategies on various topics, such as Data Science and Machine Learning.

This episode includes a re-distribution of the interview I had with Daliana Liu on her podcast "The Data Scientist Show" (https://open.spotify.com/episode/0HzWSLIWpz6iHJiKXs0GNJ) There, I shared my thoughts about the focus difference between machine learning engineering and product data science, some projects I worked on at Uber and LinkedIn, as well as other topics: such as why I transi...