---
title: "Machine Learning"
description: "622 posts about Machine Learning, summarised, each linking to the original."
---

# Machine Learning
> 622 posts about Machine Learning, summarised, each linking to the original.

## Articles

### [The evolution of Grab's machine learning feature store](https://yomu.fyi/post/the-evolution-of-grab-s-machine-learning-feature-store.md)
- Company: [Grab](https://yomu.fyi/company/grab.md)
- Author: Daniel Tai
- Published: Jul 24, 2025
- Type: Explainer

Grab describes how it evolved its machine learning feature store as increasingly complex models began requiring high-cardinality keys and high-dimensional data. It replaced its initial Amphawa approach with feature tables: data-lake tables whose contents are atomically ingested into serving storage, with emphasis on consistency and reproducibility. The workflow stores feature tables and version-control metadata as Parquet objects in Amazon S3, validates schemas, loads data through reverse ETL into Aurora Postgres, and warms new tables through shadow reads before switching. The redesign supports contextual retrieval, atomic collection-wide updates, and isolation between reading and writing capacity; its SDK remains a thin database-driver wrapper that translates queries to SQL. Grab also chose decentralised deployments for version flexibility, rollout strategies, noisy-neighbor mitigation, and improved caching, concluding that Amazon Aurora for Postgres provides a robust, scalable, cost-effective foundation.


### [Leveraging multimodal LLMs for Shopify’s global catalogue: Recap of expo talk at ICLR 2025](https://yomu.fyi/post/leveraging-multimodal-llms-for-shopify-s-global-catalogue-recap-of-exp.md)
- Company: [Shopify](https://yomu.fyi/company/shopify.md)
- Author: 2023-10-18
- Published: Jul 16, 2025
- Type: Explainer

Shopify’s Global Catalogue addresses fragmented, multilingual, multimodal product data that makes semantic search, faceting, matching, and AI-driven discovery difficult across millions of merchants. The system combines product data, understanding, matching, and reconciliation layers to standardize metadata, identify duplicate listings, and create canonical records with Universal Product IDs. Shopify fine-tunes vision large language models for multiple tasks per catalogue entity, while streaming infrastructure processes more than 10 million product updates daily and serves 40 million multimodal LLM-powered inferences. The resulting data supports merchant-admin suggestions, search, recommendations, embeddings, and conversational commerce, although scalability, accuracy, latency, model consolidation, graph reasoning, and continuous retraining remain active challenges.


### [Augmented commerce: Machine learning at Shopify](https://yomu.fyi/post/augmented-commerce-machine-learning-at-shopify.md)
- Company: [Shopify](https://yomu.fyi/company/shopify.md)
- Author: 2023-10-18
- Published: Jul 4, 2025
- Type: Explainer

Shopify frames commerce as a complex, continuous optimization problem in which merchants have different definitions of success and need support across the business lifecycle. Its machine-learning work uses a historical dataset covering millions of businesses to optimize product metadata, transaction fraud assessment, merchant assistance, product discovery, capital decisions, behavioral modeling, and search query rewriting. The systems include fine-tuned Qwen multimodal models, LLaMa models and MCPs for Sidekick, Nomic embeddings for billions of products, a tabular transformer for forecasting merchant GMV, and experiments with HSTU sequence models. To support this work, Shopify partners with GCP, Nebius, CentML, and Toloka, while providing access to GPUs, data services, frontier models, and internal tooling; the post also presents these efforts as part of its hiring pitch.


### [The Future of React Native Graphics: WebGPU, Skia, and Beyond](https://yomu.fyi/post/the-future-of-react-native-graphics-webgpu-skia-and-beyond.md)
- Company: [Shopify](https://yomu.fyi/company/shopify.md)
- Author: 2023-10-18
- Published: Jun 20, 2025
- Type: Announcement

React Native WebGPU and React Native Skia are presented as a graphics stack for advanced 2D, 3D, and general-purpose GPU workloads in native applications. The work uses Skia Graphite with Dawn, Google’s open-source WebGPU implementation, to replace separate OpenGL and Metal backend implementations with a unified architecture, while enabling Three.js, React Three Fiber, TensorFlow.js, and compute shaders. Three.js Shading Language lets the same shaders target WebGPU and WebGL, and Shopify’s Black Friday globe prototypes run natively with identical results to the web version. Architectural changes in React Native Skia moved from a mutable SkiaDOM to an immutable fabric reconciler, producing reported animation gains of up to 50% on iOS and nearly 200% on Android, plus 98% of reported Android crashes fixed. Graphite remains experimental, but adds dedicated-thread canvases and bidirectional 2D/3D composition, while Skia also supports macOS, tvOS, and headless Node.js rendering.


### [Odysseus to AI: Matt Greenwood on the Dev Interrupted Podcast](https://yomu.fyi/post/odysseus-to-ai-matt-greenwood-on-the-dev-interrupted-podcast.md)
- Company: [Two Sigma](https://yomu.fyi/company/two-sigma.md)
- Author: Emily Majewski
- Published: May 27, 2025
- Type: Explainer

Matt Greenwood, Chief Innovation Officer at Two Sigma, outlines strategies for managing technological change, integrating artificial intelligence into systematic investment, and building supportive engineering organizations. Greenwood describes sustained innovation through the S-curve using an epsilon and omega approach, combining small iterative steps with a broad long-term vision. To direct resources amid rapid advances in machine learning and large language models, he presents a functional framework categorizing AI roles into advisory insights, oracle outcome validation, operational task automation, and agentic coordination. This categorization aims to automate routine workflows while keeping human creativity, control, and higher-level thinking at the center of the investment process. Additionally, the organization fosters employee engagement through initiatives such as an internal hacker lab where cross-disciplinary teams build projects ranging from robots to racing simulators.


### [Evolution of Product Classification at Shopify: From Categories to Comprehensive Product Understanding](https://yomu.fyi/post/evolution-of-product-classification-at-shopify-from-categories-to-comp.md)
- Company: [Shopify](https://yomu.fyi/company/shopify.md)
- Author: 2023-10-18
- Published: May 8, 2025
- Type: Explainer

Shopify’s product classification system evolved from a 2018 logistic-regression model using TF-IDF into a product-understanding platform designed to capture categories, attributes, characteristics, metadata, and content-safety signals. A 2020 multimodal image-and-text approach improved classification, but growing product complexity led Shopify to combine Vision Language Models with its Standard Product Taxonomy, spanning more than 10,000 categories and over 1,000 attributes across more than 26 business verticals. The production pipeline uses two-stage category and attribute prediction, with category context shaping the second call, and runs on Kubernetes with NVIDIA GPUs; FP8 quantization, Nvidia Dynamo in-flight batching, and KV caching improve inference efficiency. A multi-LLM annotation system independently evaluates products and uses arbitration when models disagree, helping build training data for consistent predictions. The resulting system processes over 30 million predictions daily, delivers an 85% predicted-category acceptance rate, and doubles hierarchical precision and recall versus the earlier neural network approach.


### [How On-Device AI Models Find Your Best Tinder Profile Photos](https://yomu.fyi/post/how-on-device-ai-models-find-your-best-tinder-profile-photos.md)
- Company: [Tinder](https://yomu.fyi/company/tinder.md)
- Author: Tinder
- Published: Apr 28, 2025
- Type: Problem & solution

Tinder developed an on-device AI Photo Selector to recommend compelling profile photos directly from a user's camera roll without exposing private assets. To establish a reference face, the system extracts facial landmarks via Apple's Vision framework and normalizes the crop using affine transformations. To prevent funnel drop-offs from mandatory selfies, the architecture includes a fallback mechanism that detects a single recognizable face in existing profile photos. Processing workflows execute concurrently across thousands of local assets while downloading and initializing TensorFlow Lite models in parallel using Combine. A dedicated analytics queue synchronizes operation counts with DispatchGroup to ensure accurate metric capture before the global rollout.


### [Stripe’s new AI Assistant in VS Code](https://yomu.fyi/post/stripe-s-new-ai-assistant-in-vs-code.md)
- Company: [Stripe](https://yomu.fyi/company/stripe.md)
- Author: Mathew Varughese
- Published: Mar 18, 2025
- Type: Announcement

Stripe has introduced an AI Assistant in its VS Code extension to answer developer questions using current Stripe knowledge rather than relying on LLM pre-training. The extension retrieves API reference entries, integration guides, code examples, and curated summaries of developer Discord threads, while inserting the user’s API key into generated snippets and supporting either GitHub Copilot’s @stripe agent or its own chat interface. Its retrieval pipeline classifies queries, combines BM25 keyword and k-nearest-neighbor embedding search, reranks results, and sends selected sources, code, user code, and the question to Claude Sonnet through a RAG prompt. A nightly Temporal workflow refreshes indexed content, while evaluation uses golden and synthetic datasets, Mean Reciprocal Rank, and human and automated review. On its synthetic test dataset, the system includes the best source about 91.11% of the time and reports an MRR of about 78%.


### [Behind the scenes of Canva's DesignDNA campaign](https://yomu.fyi/post/behind-the-scenes-of-canva-s-designdna-campaign.md)
- Company: [Canva](https://yomu.fyi/company/canva.md)
- Author: Divya Patel
- Published: Feb 24, 2025
- Type: Problem & solution

Canva launched DesignDNA in December 2024 as a personalized year-in-review campaign to highlight user achievements and showcase generative AI capabilities. Because internal privacy rules strictly prohibited inspecting personal designs, the engineering and creative teams inferred user preferences from the style and theme metadata tags on public templates. An initial keyword-matching algorithm paired 95% of users with one of seven emerging design trends, while generative AI keyword expansion increased coverage to 99%. The team also synthesized over one million localized poems and distinct design personalities using tools like Magic Write and Dream Lab. Finally, dynamic Canva template elements populated through URL parameters assembled 95 million distinct multi-page stories across nine locales.


### [Turbo-Charging ML Development](https://yomu.fyi/post/turbo-charging-ml-development.md)
- Company: [Ramp](https://yomu.fyi/company/ramp.md)
- Author: Ryan Stevens, Ryne Carbone
- Published: Feb 10, 2025
- Type: Problem & solution

Ramp describes Turbo, a YAML-based configuration system created to reduce the time and engineering effort required to deploy machine-learning models while preserving reproducibility and explainability. Its abstraction represents end-to-end feature and model pipelines, datasets, and shared train and predict jobs; Python entities and aggregates built with Pydantic are materialized through JobSpec and executed locally or remotely before models are stored. A developer can run a training workflow with a YAML file and a few lines of code, then adapt the same schema for prediction by changing a few configuration lines. The first Credit Risk model required nearly four months and more than 3,000 lines of deployment code; with Turbo, new models deploy in under five minutes using about 40 YAML lines on average. The system also standardizes practices such as holiday alignment in time-series models and supports nearly 20 deployed models built by nine developers.


### [AI Core Team Lead Mike Schuster on How to Get the Most From LLMs](https://yomu.fyi/post/ai-core-team-lead-mike-schuster-on-how-to-get-the-most-from-llms.md)
- Company: [Two Sigma](https://yomu.fyi/company/two-sigma.md)
- Author: Joy Looney
- Published: Feb 4, 2025
- Type: Explainer

Mike Schuster, Head of the AI Core Team at Two Sigma, advocates for grounding large language model adoption in practical tasks rather than speculative industry hype. Realistic enterprise applications focus on accelerating data processing, running faster experiments, and extracting domain-specific features from transcripts such as earnings calls and Federal Reserve speeches via prompt engineering. Because financial data faces inherent volume limits across trading days, successful deployments require multidisciplinary human teams to balance rapid technical experimentation with rigorous domain expertise and analytical reasoning. Schuster also dismisses predictions that programming will become obsolete, comparing coding to learning a musical instrument that cultivates structured thinking, problem decomposition, and scientific common sense essential for building complex predictive models.


### [Image replacement in Canva designs using reverse image search](https://yomu.fyi/post/image-replacement-in-canva-designs-using-reverse-image-search.md)
- Company: [Canva](https://yomu.fyi/company/canva.md)
- Author: Sam Jacobs
- Published: Jan 28, 2025
- Type: Problem & solution

Canva needed an automated way to replace media in design templates, such as when third-party licensing partnerships expire across more than 150 million images. Existing recommendation engines, perceptual hashing, and text metadata searches failed to capture visual similarity hierarchies or ensure replacement relevance. To build a reverse image search system, engineers evaluated embedding models including CLIP, ViTMAE, DreamSim, CaiT, and DINOv2 alongside an external vector database supporting metadata filtering. Evaluation on sample datasets identified DINOv2 as the best model for preserving subjects, background context, and color tones in photos. Integrated into the Template Assistant as a human-in-the-loop tool, the automated suggestions increased image replacement speeds by 4.5 times during initial pilot testing.


### [Scaling Large Language Models for e-Commerce: The Development of a Llama-Based Customized LLM](https://yomu.fyi/post/scaling-large-language-models-for-e-commerce-the-development-of-a-llam.md)
- Company: [Ebay](https://yomu.fyi/company/ebay.md)
- Author: Christian Herold and Shahram Khadivi
- Published: Jan 17, 2025
- Type: Problem & solution

eBay developed e-Llama, customized 8B and 70B language models, by adapting Meta’s Llama 3.1 models to e-commerce rather than training entirely from scratch. The approach uses continued pretraining on filtered public listings, product reviews, classifier-selected open-source examples, replay data, and 10% non-English general-domain data to add domain knowledge while limiting catastrophic forgetting. Training ran on 480 NVIDIA H100 GPUs using Megatron-LM, 3D parallelism, flash-attention-2, a 1:1 general-to-e-commerce sampling ratio, a learning-rate maximum set to 10% of the original, and one trillion tokens. The 70B model required about one month and 340,000 GPU-hours. E-commerce benchmarks improved approximately 25% in English and 30% in non-English, while general-domain NLU degraded 1% for e-Llama 70B; subsequent instruction tuning added safety and instruction-following guardrails.


### [How Shopify improved consumer search intent with real-time ML](https://yomu.fyi/post/how-shopify-improved-consumer-search-intent-with-real-time-ml.md)
- Company: [Shopify](https://yomu.fyi/company/shopify.md)
- Author: 2023-10-18
- Published: Oct 18, 2024
- Type: Explainer

Shopify describes how it built near-real-time text and image embedding pipelines to support semantic search, replacing keyword matching with representations that enable similarity-based retrieval. Using Google Cloud’s Dataflow and Apache Beam, the pipelines process roughly 2,500 embeddings per second, preprocess images, run GPU-backed inference, and write outputs to both a data warehouse and an event topic for downstream ingestion. The engineering discussion focuses on trade-offs among memory, throughput, latency, and cost: reducing harness threads cut memory use about 2.6x and allowed a return to cheaper workers, while model sharing reduced throughput. For batching, bursty input produced bundles of one, and stateful batching was rejected because its shuffle increased latency; Dataflow’s per-process model instances instead provided enough parallelism to saturate GPUs.


### [What is a Vector Database?](https://yomu.fyi/post/what-is-a-vector-database.md)
- Company: [Databricks](https://yomu.fyi/company/databricks.md)
- Author: Databricks Staff
- Published: Oct 11, 2024
- Type: Explainer

Vector databases store and manage high-dimensional vector representations of data, grouping them by similarity rather than traditional row-and-column organization. They support semantic and similarity-based retrieval by converting text, images, audio, or video into embeddings, indexing them with methods such as product quantization and HNSW, and comparing query vectors with indexed vectors. These capabilities underpin use cases including RAG, hybrid search, recommendations, NLP, image and video recognition, and anomaly detection, while sharding, partitioning, caching, and replication support scale. The article distinguishes a vector database from a standalone index because it also provides storage, CRUD, metadata filtering, horizontal scaling, and data management functions, and describes Databricks AI Search as an integrated implementation with automatic scaling, optimization, security, and governance.


### [Evolution of Catwalk: Model serving platform at Grab](https://yomu.fyi/post/evolution-of-catwalk-model-serving-platform-at-grab.md)
- Company: [Grab](https://yomu.fyi/company/grab.md)
- Author: Vishal Sharma
- Published: Oct 1, 2024
- Type: Problem & solution

Grab developed and scaled Catwalk, an internal machine learning model serving platform, to address operational bottlenecks, low resource utilization, and deployment friction between data scientists and backend engineers. The platform transitioned from an admin-managed TensorFlow Serving setup into a low-code self-service system supporting PyTorch and ONNX, before replacing complex Helm charts with Kubernetes Custom Resource Definitions for declarative, blue-green deployment orchestration. To support complex business workflows and multi-model applications, Grab subsequently introduced Catwalk Orchestrator with bundled deployments that allow individual services to scale independently. Across two years, the orchestrator architecture expanded to 200 deployed applications serving approximately 1,400 production machine learning models.


### [AI GPU Clusters, From Your Laptop, With Livebook](https://yomu.fyi/post/ai-gpu-clusters-from-your-laptop-with-livebook.md)
- Company: [Fly.io](https://yomu.fyi/company/fly-io.md)
- Published: Sep 24, 2024
- Type: Explainer

Livebook, FLAME, and the Nx stack combine Elixir notebooks, elastic execution, and GPU-backed machine learning into a workflow that can move from a laptop to remote infrastructure. Livebook can start runtimes on Fly Machines, connect to existing Erlang/Elixir applications, and synchronize notebook dependencies and modules across nodes while preserving access to private resources within a Fly organization. FLAME wraps arbitrary code in Flame.call, manages executor pools, controls node limits and concurrency, streams results, and shuts down idle nodes or the cluster after disconnection. In the demonstrated AI workflow, video stills are processed on GPU Fly Machines with Llama, then Mistral summarizes the returned descriptions. The same setup also drives a 64-GPU hyperparameter run for BERT, with each L40s node preparing dependencies, loading data, fine-tuning, and streaming chart results to Livebook.


### [Chimera Sandbox: A scalable experimentation and development platform for Notebook services](https://yomu.fyi/post/chimera-sandbox-a-scalable-experimentation-and-development-platform-fo.md)
- Company: [Grab](https://yomu.fyi/company/grab.md)
- Author: Arkar Min Aung
- Published: Aug 27, 2024
- Type: Announcement

Grab's Chimera team built Chimera Sandbox to address resource accessibility challenges and enable rapid machine learning experimentation across technical and non-technical teams. The platform runs on multi-cluster Kubernetes infrastructure capable of scaling to thousands of nodes per cluster during peak loads. Users can provision customizable Docker environments with varying CPU and GPU configurations, collaborate in real time, and query data using Spark-on-Kubernetes integrations and helper magic cells. The workspace integrates with Grab's AI Gateway for LLM access alongside tools like MLflow, Optuna, RStudio, and LabelStudio on shared storage, backed by user-level cost attribution.


### [Improving Retrieval on Ramp with Transaction Embeddings](https://yomu.fyi/post/improving-retrieval-on-ramp-with-transaction-embeddings.md)
- Company: [Ramp](https://yomu.fyi/company/ramp.md)
- Author: Calix Huang, Anton Biryukov
- Published: Aug 13, 2024
- Type: Problem & solution

Ramp developed transaction embeddings to help automate accounting coding, where employees choose general ledger (GL) categories for transactions. The model represents enriched transaction features—such as merchant, category, department, location, amount, memo, spend program, and trip—as stringified documents labeled with chart-of-accounts codes. Starting from a pre-trained encoder, it is fine-tuned with sentence-transformers and BatchSemiHardTripletLoss, using large batches and contextual sampling to learn embeddings whose similarity is measured with cosine similarity. Triplet loss pulls same-label transactions together and pushes different-label examples apart, while informed mining addresses subtle distinctions such as Travel: Sales versus Travel: Engineering. The embeddings support GL coding suggestions, high-confidence defaults, business-spend similarity analysis, and transaction context for LLM-enabled features, with the stated aim of keeping predictions personalized and data private.


### [Machine Learning in Content Moderation at Etsy](https://yomu.fyi/post/machine-learning-in-content-moderation-at-etsy.md)
- Company: [Etsy](https://yomu.fyi/company/etsy.md)
- Author: David Azcona
- Published: Aug 6, 2024
- Type: Explainer

Etsy describes a supervised-learning system for detecting potentially violating listings at marketplace scale, alongside community reporting and other automated controls. Training data combines logged and annotated positives, hard negatives, and randomly sampled easy negatives, with multimodal textual and imagery signals split by time for progressive evaluation. The architecture uses BERT-based text encoding with ALBERT for reduced compute, EfficientNet image encoding, late concatenation of modality representations, a softmax output, and focal loss for class imbalance. Etsy evaluates candidates offline and in production through API-serving, canary releases, and champion-challenger A/B tests, while using proxy recall because full manual review is infeasible. Frequent retraining and monitoring support ongoing iteration; the methodology has identified and removed more than 100,000 violations, alongside other tools and services.


[Newer posts](https://yomu.fyi/topic/machine-learning/page/22.md) · [Older posts](https://yomu.fyi/topic/machine-learning/page/24.md)