# Yelp News from the Yelp Engineering and Product Teams This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## ML based ranking using Nrtsearch DevFeed: [ML based ranking using Nrtsearch]() Original publisher: [Read original article]() Author: Mohammad Mohtasham (Software Engineer); Tao Yu (Software Engineer) Published: 2026-09-16T00:00:00Z Content type: article Language: en Sources: [Yelp]() Topics: [Inference](), [Scalability](), [Latency](), [recommendation systems](), [Open Source]() Tags: [bridge](), [inference](), [machine-learning](), [overhead](), [performance](), [pipeline](), [scalability](), [service]() ### AI overview Yelp's Nrtsearch Inference Plugin embeds machine-learning ranking directly in the search layer. The article explains the scoring workflow, including model configuration, feature extraction, candidate ranking, and application-specific business logic. It describes how co-locating feature storage and inference reduces network transfer, serialization overhead, and latency compared with a standalone inference service. ### Source excerpt We've extended Nrtsearch with the Inference Plugin, which embeds ML-based ranking directly in the search layer -- eliminating the need for a standalone scoring service. We use Nrtsearch (read more information on the blog post), a Lucene-based open-source search engine built by Yelp, to power a variety of applications such as business search, reviews search, ad delivery and photo search. In this blog post, we give a high-level overview of the Machine Learning (ML) based scoring workflow in Nrtsearch. We'll show how ML models are configured and loaded, and how different applications use custom business logic to develop, test, and... ## Building Menu Vision: Real-Time Dish Recognition DevFeed: [Building Menu Vision: Real-Time Dish Recognition]() Original publisher: [Read original article]() Author: Arpitha Dudi, Growth Tech Lead Published: 2026-08-20T00:00:00Z Content type: article Language: en Sources: [Yelp]() Topics: [Computer vision](), [Artificial Intelligence](), [Android](), [cameraX](), [ML Kit](), [Hackathon](), [Development](), [iOS]() Tags: [ai](), [android](), [camerax](), [data-pipeline](), [development](), [hackathon](), [ios](), [ml-kit](), [pipeline](), [prototype](), [real-time](), [recognition](), [vision]() ### AI overview Yelp describes Menu Vision, a feature that uses a phone camera, AI, augmented reality, and text recognition to identify dishes on restaurant menus and surface related user photos and reviews. The article covers its hackathon-origin Android prototype and the production system's on-device machine learning, real-time computer vision, fuzzy matching, and distributed data pipeline. ### Source excerpt Menus aren't just lists, they're a window into a restaurant's unique offerings, specialties, and personality, shaping where and what we choose to eat. But here's the challenge: reading "Kung Pao Chicken - Stir-fried chicken with peanuts in spicy sauce" doesn't tell you what the portions look like, whether other diners loved it, or if it matches your expectations. At Yelp, we knew we had the solution sitting in our user-generated content: hundreds of millions of photos, reviews, and prices for dishes. The problem? Users had to manually search for each dish, an experience that doesn't work well when you're at... ## Migrating a Large Flow Monorepo to TypeScript DevFeed: [Migrating a Large Flow Monorepo to TypeScript]() Original publisher: [Read original article]() Author: Shawn Walton, Software Engineer Published: 2026-08-04T00:00:00Z Content type: article Language: en Sources: [Yelp]() Topics: [monorepo](), [TypeScript](), [migration](), [repo](), [Code]() Tags: [coverage](), [integrity](), [migration](), [monorepo](), [open-source](), [rewrite](), [typescript]() ### AI overview Yelp describes its gradual migration of a large Flow monorepo to TypeScript. The team chose a package-by-package approach to preserve type-checking trust and type safety while Flow and TypeScript coexisted in production, using open-source conversion tooling, custom codemods, and manual fixes. ### Source excerpt In early 2017, Webcore selected Flow as Yelp's next-generation typechecker over TypeScript. At the time there was no clear frontrunner. Flow had better support for React, better performance, and a respectable repository of types for NPM packages in flow-typed. However, a few years later the landscape had shifted, and it became clear that TypeScript was emerging as the stronger choice. TypeScript's DefinitelyTyped repo had thousands more packages than flow-typed, and many more packages shipped with built-in TypeScript headers. TypeScript also proved to be more stable over time, and limited uptake in the wider JS community made finding tools and documentation... ## Migrating from Apollo Tooling to GraphQL Codegen at Yelp DevFeed: [Migrating from Apollo Tooling to GraphQL Codegen at Yelp]() Original publisher: [Read original article]() Author: Igor Kusakov, Software Engineer Published: 2026-07-16T00:00:00Z Content type: article Language: en Sources: [Yelp]() Topics: [GraphQL](), [Code generation](), [migration](), [Tooling](), [React](), [TypeScript](), [monorepo]() Tags: [code-generation](), [deprecated](), [graphql](), [migrate](), [monorepo](), [react](), [tooling](), [typescript]() ### AI overview Yelp describes migrating its frontend React monorepo from deprecated Apollo Tooling to GraphQL Codegen for generating TypeScript definitions from GraphQL queries and the server schema. ### Source excerpt Introduction At Yelp, we rely heavily on GraphQL and Apollo for data loading in our frontend React monorepo. When a developer writes a GraphQL query or mutation inside a React component, the shape of the response is defined by the query itself -- but TypeScript has no way to know what that shape looks like at compile time. That's where code generation (codegen) comes in. We rely on codegen to read our queries and server schema, then produce TypeScript type definitions for every operation's input variables and output data. In our codebase, this looks like: src/components/UserProfile/ ├── UserProfile.tsx <- contains... ## Training Orchestrator: Unifying Model Training at Yelp DevFeed: [Training Orchestrator: Unifying Model Training at Yelp]() Original publisher: [Read original article]() Author: Ying Wang and Nathan Sponberg, Software Engineer Published: 2026-07-14T00:00:00Z Content type: article Language: en Sources: [Yelp]() Topics: [Machine Learning & Artificial Intelligence](), [Orchestration](), [Apache Spark](), [Training AI Models](), [Model Development](), [configuration](), [configuration-management](), [Monitoring]() Tags: [configuration](), [machine-learning](), [maintenance](), [mlflow](), [model-training](), [orchestration](), [reproducibility](), [spark](), [tooling]() ### AI overview Yelp's Core Machine Learning Team developed Training Orchestrator to standardize how machine learning teams define and run Spark-based model training. The configuration-driven system addresses duplicated code, inconsistent configurations, limited local testing, scattered validation and monitoring, and poor reproducibility across environments. ### Source excerpt At Yelp, we train many machine learning models on different schedules. Applied machine learning teams all have their own set of Spark-based training batches, scripts, and configurations. Over time, these diverged, leading to duplicated code, subtle inconsistencies, and a growing maintenance burden. Yelp's Core Machine Learning Team has developed excellent tooling across our ML ecosystem over the years: feature stores for reproducible data, a unified training library for neural networks and gradient-boosted trees, seamless Spark integration, and MLflow services for model tracking and deployment. But there was still one key piece missing right in the middle: a standardized way to... ## Beyond the Menu Tree: How Yelp Built a Smarter Customer Success Chatbot with AI DevFeed: [Beyond the Menu Tree: How Yelp Built a Smarter Customer Success Chatbot with AI]() Original publisher: [Read original article]() Author: Lina Lee, Machine Learning Engineer; Nelson Lee, Engineering Manager Published: 2026-05-27T00:00:00Z Content type: article Language: en Sources: [Yelp]() Topics: [Retrieval Augmented Generation (RAG)](), [Support](), [AI Chat](), [Large Language Model](), [AI Development]() Tags: [ai](), [architecture](), [customer](), [deep-dive](), [engineering](), [large-language-model](), [legacy](), [llm](), [rag](), [retrieval](), [support](), [technical](), [workflow]() ### AI overview Yelp describes replacing its legacy Customer Success chatbot's fixed menu and phrase matching with an LLM-assisted system connected to an internal knowledge base through a Retrieval Augmented Generation pipeline. The system routes customer queries into specialized workflows, including question answering, refunds, and billing. ### Source excerpt The Evolution of Support: From Fixed Phrases to Conversation At Yelp, delivering responsive and accurate customer support is a core priority. For years, our legacy Customer Success (CS) Chatbot provided support by guiding users through a static support experience. Users either navigated a 2-step menu tree or typed a query that was matched against a fixed set of phrases to retrieve an answer. While functional, the legacy chatbot had a key limitation: its reliance on rigid matching meant that if a query didn't fit the menu structure or precisely match a known phrase, the user wouldn't be able to get... ## How Partition Access Visualizations Reduced our Data Lake S3 Cost by 33% DevFeed: [How Partition Access Visualizations Reduced our Data Lake S3 Cost by 33%]() Original publisher: [Read original article]() Author: Nick Del Nano, Data Streaming Published: 2026-05-21T00:00:00Z Content type: article Language: en Sources: [Yelp]() Topics: [data](), [data lake](), [Apache Iceberg](), [Amazon S3](), [AWS IAM]() Tags: [analytics](), [apache-iceberg](), [aws](), [data](), [data-lake](), [iam](), [partition](), [s3]() ### AI overview Yelp describes visualizations that map partition keys against access-event timestamps to reveal daily batch jobs, backfills, and ad hoc queries. The resulting usage attribution supported Apache Iceberg migration and storage-efficiency work that reduced the cost of its petabyte-scale data lake by 33%. ### Source excerpt Introduction In large analytics environments, data teams often struggle to answer deceptively simple questions, like who their stakeholders are and how their data is being used. At Yelp, we address this by visualizing access patterns, plotting time-based partition key values against access event timestamps. These visualizations reveal distinct usage signatures - ad hoc queries, daily batch jobs, and periodic backfills - allowing data owners to understand their stakeholders and use cases. This deeper insight into data usage has enabled high-impact platform initiatives including migrating thousands of tables to Apache Iceberg format and identifying storage efficiencies which reduced the cost of... ## Optimizing Our Build Times by Migrating from Webpack to Rspack DevFeed: [Optimizing Our Build Times by Migrating from Webpack to Rspack]() Original publisher: [Read original article]() Author: Benson Pan, Software Engineer Published: 2026-05-20T00:00:00Z Content type: tutorial Language: en Sources: [Yelp]() Topics: [Webpack](), [build times](), [monorepo](), [build performance](), [Developer experience]() Tags: [build](), [build-times](), [build-tools](), [developer-experience](), [migration-guide](), [monorepo](), [performance](), [speed](), [tooling](), [tools](), [webpack]() ### AI overview A Yelp engineering blog post explains how the team migrated monorepo builds from Webpack to Rspack. It reports an approximately 50% reduction in build time and discusses compatibility, configuration, custom plugins, and a staged rollout. ### Source excerpt Over the years, Webpack has remained the bundler of choice for many JS projects, including here at Yelp. While it has served us well, its speed has increasingly become a bottleneck as our monorepo continues to grow. Fortunately, a bunch of new build tools (Vite, Parcel, Rspack, etc.) have emerged in recent years. Each of these tools promises different ways of improving performance and developer experience. In this blog post, we'll walk through how we migrated our monorepo builds from Webpack to Rspack and achieved an approximately 50% reduction in build time. Why Rspack Our team has been closely observing... ## ML based ranking using Nrtsearch DevFeed: [ML based ranking using Nrtsearch]() Original publisher: [Read original article]() Author: Mohammad Mohtasham (Software Engineer); Tao Yu (Software Engineer) Published: 2026-05-11T00:00:00Z Content type: tutorial Language: en Sources: [Yelp]() Topics: [Machine Learning & Artificial Intelligence](), [Inference](), [AI Inference](), [model-deployment](), [backends](), [Open Source]() Tags: [blog-post](), [inference](), [latency](), [machine-learning](), [ml](), [models](), [network](), [open-source](), [performance](), [pipeline](), [ranking](), [software]() ### AI overview Yelp extended its Lucene-based Nrtsearch engine with an Inference Plugin that embeds machine-learning ranking directly in the search layer. The article explains the ranking workflow, including model configuration and loading, and describes how co-locating feature storage and inference reduces network transfer, serialization overhead, and latency compared with a standalone inference service. ### Source excerpt We've extended Nrtsearch with the Inference Plugin, which embeds ML-based ranking directly in the search layer -- eliminating the need for a standalone scoring service. We use Nrtsearch (read more information on the blog post), a Lucene-based open-source search engine built by Yelp, to power a variety of applications such as business search, reviews search, ad delivery and photo search. In this blog post, we give a high-level overview of the Machine Learning (ML) based scoring workflow in Nrtsearch. We'll show how ML models are configured and loaded, and how different applications use custom business logic to develop, test, and... ## How Yelp Keeps Server-Driven UI Consistent Across Four Platforms DevFeed: [How Yelp Keeps Server-Driven UI Consistent Across Four Platforms]() Original publisher: [Read original article]() Author: Radu Comaneci, Software Engineer Published: 2026-04-22T00:00:00Z Content type: article Language: en Sources: [Yelp]() Topics: [Design system](), [cross-platform](), [ui](), [Web](), [Android](), [iOS](), [Framework]() Tags: [android](), [cross-platform](), [design-system](), [ios](), [platforms](), [server-driven-ui](), [ui](), [web]() ### AI overview This article explains how Yelp integrates its CHAOS server-driven UI framework with the Cookbook design system and the auto-generated Konbini bridge library. The approach is intended to keep UI components and their behavior consistent across Yelp's Web, iOS, and Android applications, including Python/CHAOS backend representations. ### Source excerpt If you've read our earlier post, you already know about CHAOS--the server-driven UI (SDUI) framework we built at Yelp that powers our dynamic views. Until now, we've explored its architecture, backend implementation, and component model. In this post, we'll dive into how we integrated CHAOS with Yelp's cross-platform design system, Cookbook, and the auto-generated bridge library, Konbini. Introduction to Cookbook At Yelp, we support two major applications across our Web, iOS, and Android platforms: Yelp and Yelp for Business. This results in six different variations, which makes it challenging to maintain a unified experience. To address this challenge, we created... ## Zero downtime Upgrade: Yelp's Cassandra 4.x Upgrade Story DevFeed: [Zero downtime Upgrade: Yelp's Cassandra 4.x Upgrade Story]() Original publisher: [Read original article]() Author: Mark Surnin and Muhammad Junaid Muzammil, Software Engineer Published: 2026-04-07T00:00:00Z Content type: article Language: en Sources: [Yelp]() Topics: [Apache Cassandra](), [upgrade](), [NoSQL](), [Database](), [observability](), [Kubernetes]() Tags: [2026](), [apache](), [cassandra](), [clusters](), [downtime](), [kubernetes](), [nosql](), [observability](), [reliability-engineering](), [upgrade]() ### AI overview Yelp's Database Reliability Engineering team describes upgrading more than a thousand Cassandra nodes from 3.11 to 4.1 on Kubernetes without downtime. The article covers the motivation, expected reliability and performance improvements, operational guardrails, certificate handling, repairs, logging, and compatibility work for related components. ### Source excerpt The Database Reliability Engineering team at Yelp seamlessly upgraded more than a thousand Cassandra nodes with zero downtime. This post takes you behind the scenes of our upgrade strategy, from planning sessions to flawless rollouts. Background Motivation Apache Cassandra is a distributed wide-column NoSQL datastore and is used widely at Yelp for storing both primary and derived data. Yelp orchestrates Cassandra clusters on Kubernetes with the help of operators, as explained in our operator overview post. Upgrading from Cassandra 3.11 to 4.1 offered several observability and reliability improvements, in addition to performance gains. Based on public benchmarks, we expected to...