---
title: "Latest reads"
description: "The engineering internet, summarised so you can actually read it."
---

# Latest reads
> The engineering internet, summarised so you can actually read it.

## 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

Grab redesigned its initial machine learning feature store, Amphawa, to address high-dimensional data, complex entity retrieval, and versioning challenges during feature updates. The new architecture adopts a feature-table model where data scientists output Parquet datasets to Amazon S3 using Spark, which are then atomically ingested into Amazon Aurora PostgreSQL via a reverse ETL workflow. To prevent noisy-neighbor contention and optimize infrastructure costs, the platform utilizes Aurora's distributed storage to separate reads from writes. Grab pairs Aurora Serverless on writer nodes to scale up during daily batch ingestion with Provisioned instances on read replicas for steady serving traffic.


### [Workflows: Automatically customize an object with metadata](https://yomu.fyi/post/workflows-automatically-customize-an-object-with-metadata.md)
- Company: [Stripe](https://yomu.fyi/company/stripe.md)
- Author: Ashley Ansari
- Published: Jul 22, 2025

Stripe Workflows can customize Stripe objects with metadata, adding contextual key-value information without changing the core data model or storing it separately. The example targets customers associated with successful Payment Intents that include a tip, so a customer can be labeled for business use, such as reconciliation, reporting, or internal integrations. To build it, the workflow retrieves the customer from the Payment Intent, emails a team member, checks whether the successful Payment Intent contains a non-empty tip amount, and updates that customer with the metadata key Tier and value VIP. Workflows runs a trigger and sequential actions or conditions in a visual Stripe Dashboard builder, while metadata can carry information to later workflow steps and support automation, analytics, and debugging.


### [Grab's service mesh evolution: From Consul to Istio](https://yomu.fyi/post/grab-s-service-mesh-evolution-from-consul-to-istio.md)
- Company: [Grab](https://yomu.fyi/company/grab.md)
- Author: Hilman Kurniawan
- Published: Jul 16, 2025

Grab operated over 1,000 microservices across hybrid infrastructure using Consul alongside a fallback mechanism called Catcher. Single-point-of-failure vulnerabilities in Consul servers and limited support for multi-cluster operations prompted an evaluation of alternative mesh technologies, ultimately leading to the selection of Istio. Grab avoided the standard single-control-plane-per-cluster pattern by deploying multiple external control planes in dedicated Kubernetes clusters arranged in active-active pairs. Migration began in Q4 2024, shifting traffic across AWS and GCP while handling both HTTP and gRPC protocols with gradual traffic-shifting and rollback mechanisms.


### [How we built it: Jurisdiction resolution for Stripe Tax](https://yomu.fyi/post/how-we-built-it-jurisdiction-resolution-for-stripe-tax.md)
- Company: [Stripe](https://yomu.fyi/company/stripe.md)
- Author: Erich Rentz
- Published: Jul 10, 2025

Stripe Tax’s jurisdiction resolution system (JRS) determines which US taxing jurisdictions apply to a transaction, a difficult task because more than 16,000 combinations of rates and rules depend on intricate, changing boundaries. Its offline geographic information system cleans and standardizes boundary data, overlays states, counties, cities, and districts, and generates time-aware Stripe places of taxation (SPOTs), while the online system matches an address to the relevant SPOT. To reduce point-in-polygon latency, JRS indexes nested bounding boxes in a balanced R-tree rebuilt with the Sort-Tile-Recursive algorithm, narrowing candidates before applying the final polygon calculation; disjoint SPOTs are split and smaller boxes prioritized. Most states achieve address matching in a few milliseconds, with 95th-percentile latency below 10 milliseconds except South Carolina, while historical SPOT data creates an ongoing memory challenge.


### [Workflows: Creating early fraud alerts for streamlined refunds](https://yomu.fyi/post/workflows-creating-early-fraud-alerts-for-streamlined-refunds.md)
- Company: [Stripe](https://yomu.fyi/company/stripe.md)
- Author: Ashley Ansari
- Published: Jul 10, 2025

Stripe Workflows can automate responses to early fraud warnings so potentially fraudulent charges receive refunds when disputing them would be more costly. The workflow starts when an early fraud warning is created in Radar, retrieves the related charge using its Charge ID, and checks whether the amount is less than 15 USD. If the condition is met, it creates a refund; if it is not met, an optional branch can email a team member for manual review. Workflows uses a visual builder in the Stripe dashboard, where triggers, actions, and conditions run sequentially across multiple Stripe products, while Stripe Radar provides real-time fraud protection and Radar for Fraud Teams adds customization and deeper insights.


### [DispatchGym: Grab’s reinforcement learning research framework](https://yomu.fyi/post/dispatchgym-grab-s-reinforcement-learning-research-framework.md)
- Company: [Grab](https://yomu.fyi/company/grab.md)
- Author: Tan Sien Yi
- Published: Jul 7, 2025

Applying reinforcement learning to dispatch systems is often hindered when the chosen control levers exert weak influence over reward functions. To streamline research, Grab built DispatchGym, a framework that connects reinforcement learning algorithms to a dispatch process simulation via the Gymnasium API. The simulation emphasizes directional accuracy over absolute precision, allowing researchers to evaluate relative metric shifts across supply and demand scenarios. Built in modular Python and accelerated with Numba, the system allows data scientists to test code locally and launch distributed Spark executions with a single command-line call. The framework has been used to evaluate various contextual bandit models and action sampling strategies for tuning dispatch hyperparameters.


### [Stay within limits: API rate-limit-friendly pattern for Stripe webhooks](https://yomu.fyi/post/stay-within-limits-api-rate-limit-friendly-pattern-for-stripe-webhooks.md)
- Company: [Stripe](https://yomu.fyi/company/stripe.md)
- Author: Phil Leggetter
- Published: Jul 3, 2025

Stripe webhook handlers that treat events as signals and fetch the latest resource can preserve correctness when payloads are stale, partial, duplicated, or out of order, but bursts of events can drive excessive API traffic. The post describes Stripe’s general limit as 100 read requests per second and notes that exceeding it produces 429 responses. Its solution places Hookdeck Event Gateway between Stripe and the application: Hookdeck authenticates and queues incoming webhooks, throttles delivery, and lets the handler retrieve the current Stripe resource at a controlled pace. The Express.js flow verifies the Hookdeck signature, checks the event and resource ID, fetches the invoice with the Stripe SDK, and supports queue monitoring, alerts, and retries for backpressure.


### [Stripe for marketplaces: Mapping commercial relationships in code](https://yomu.fyi/post/stripe-for-marketplaces-mapping-commercial-relationships-in-code.md)
- Company: [Stripe](https://yomu.fyi/company/stripe.md)
- Author: Ana Andres
- Published: Jul 3, 2025

The post presents a guide to mapping marketplace relationships among buyers, sellers, and the platform in Stripe Connect, focusing on payment processing, seller onboarding, payouts, fees, refunds, and disputes. It uses Greens & Dairy Mart, a fictitious farm marketplace, to show how an unlicensed marketplace can manage payments with Stripe while meeting applicable regulatory requirements. Using the Account API, connected accounts, controller settings, KYC links, capabilities, Payment Intents, transfers, and webhooks, the implementation encodes responsibilities and money movements. Greens & Dairy Mart charges customers after farmers confirm dispatch, pays farmers after delivery confirmation, and retains 15% of the final price. The guide also shows refund and transfer reversal flows, including recovering disputed amounts from a seller’s balance, and concludes that Connect APIs can automate complex marketplace contracts and financial interactions.


### [Using Connect embedded components to streamline your Connect onboarding flow](https://yomu.fyi/post/using-connect-embedded-components-to-streamline-your-connect-onboardin.md)
- Company: [Stripe](https://yomu.fyi/company/stripe.md)
- Author: Jorge Aguirre Gonzalez
- Published: Jul 2, 2025

Furever, a test Connect platform for pet grooming, needs to onboard connected accounts and keep them compliant and activated while reducing friction. It first creates Custom accounts with card\_payments and transfers requested, then contrasts Stripe-hosted account links with an embedded flow built from Account Sessions, loadConnectAndInitialize, and ConnectAccountOnboarding. The integration supports appearance variables for branding, disable\_stripe\_user\_authentication when the platform owns losses and requirement collection, and onStepChange analytics for onboarding progress. A shared connectInstance can also render ConnectPayments, which provides payment listing, refunds, and dispute management without building that interface; the source also notes localization, framework support beyond React, and additional components.


### [Implementing scalable metered billing with Stripe: How Edgee handles billions of events](https://yomu.fyi/post/implementing-scalable-metered-billing-with-stripe-how-edgee-handles-bi.md)
- Company: [Stripe](https://yomu.fyi/company/stripe.md)
- Author: Alex Casalboni
- Published: Jun 24, 2025

Edgee needed usage-based billing for a business combining flat-rate subscriptions with charges for proxy requests and delivered data events, while processing billions of monthly requests across more than 100 edge locations. Its architecture captures web data at the edge through a proxy and records raw metering data centrally in Google BigQuery. To control several terabytes of data, Edgee aggregates usage by tenant and sends hourly Stripe meter events through AWS Lambda, with graduated pricing and idempotency handled by Stripe. A Go configuration defines meters, prices, flat-rate plans, and pricing tiers, while retry logic and idempotent updates support resilience during disruptions. The resulting billing system remains stateless and low-latency while providing accurate monthly invoicing and a simple hosted customer interface.


### [Counter Service: How we rewrote it in Rust](https://yomu.fyi/post/counter-service-how-we-rewrote-it-in-rust.md)
- Company: [Grab](https://yomu.fyi/company/grab.md)
- Author: Jia Long Loh
- Published: Jun 20, 2025

The Integrity Data Platform team rewrote Counter Service, a high-throughput Golang microservice serving event counts for fraud rules and machine learning models, to evaluate the operational return on investment of Rust. Rather than performing a line-by-line translation, engineers approached the service as a black box, reimplementing core application logic from scratch to satisfy established gRPC contracts across Scylla and Redis. The team resolved internal Go tooling dependencies by building custom configuration template parsers using the nom parser combinator and selected targeted open-source crates such as fred.rs and Cadence. Adapting to Rust required navigating cooperative, stackless async execution compared to Go's preemptive concurrency model, alongside managing borrow checker constraints. Ultimately, the rewrite achieved a 70% reduction in infrastructure costs while maintaining comparable service performance.


### [Measuring Commercial Impact at Scale at Canva](https://yomu.fyi/post/measuring-commercial-impact-at-scale-at-canva.md)
- Company: [Canva](https://yomu.fyi/company/canva.md)
- Author: Jun Ye
- Published: Jun 20, 2025

Canva needed a scalable, standardized way to calculate the commercial impact of thousands of annual experiments on key business metrics like Monthly Active Users and Annual Recurring Revenue. Previously, disparate teams spent over six hours per experiment performing manual, error-prone calculations across fragmented spreadsheets and inconsistent data models. To resolve this, Canva built the IMPACT app using Snowflake, Streamlit, Snowpark, and Cortex to provide a self-serve platform tied directly to its central finance model. The application scales local uplift by actual audience exposure and supports pre-experiment scenario modeling alongside post-experiment tracking. A custom deployment workflow generates pull-request-isolated Streamlit environments in Snowflake stages, reducing time-to-insight to under ten minutes while enabling multiple developers to safely build and demo features in parallel.


### [The complete stream processing journey on FlinkSQL](https://yomu.fyi/post/the-complete-stream-processing-journey-on-flinksql.md)
- Company: [Grab](https://yomu.fyi/company/grab.md)
- Author: Calvin Tran
- Published: Jun 12, 2025

Grab previously relied on Apache Zeppelin notebooks for interactive stream processing exploration, but faced lagging Flink version upgrades, five-minute cluster cold starts, and poor integration with internal platforms. To address these limitations, the team migrated to a shared FlinkSQL gateway architecture structured into compute, integration, and query layers. The new setup uses a Hive Metastore catalog to expose Kafka topics as relational tables, while a custom control plane handles authentication and headless REST APIs over Flink's native interface. For production workflows, a configuration-based portal accepts SQL logic and automatically provisions and deploys Flink pipelines within ten minutes. This transition reduced ad-hoc query response times to under one minute and eliminated the need to maintain version adapter shims.


### [Stop juggling multiple POS devices with Stripe Terminal](https://yomu.fyi/post/stop-juggling-multiple-pos-devices-with-stripe-terminal.md)
- Company: [Stripe](https://yomu.fyi/company/stripe.md)
- Author: Andrew Robinson
- Published: Jun 12, 2025

Traditional point-of-sale workflows often collect signatures, contact details, receipt preferences, and other customer information through separate devices, verbal input, or manual entry, creating errors, delays, and clutter. Stripe Terminal’s on-screen input collection lets businesses request this data on a compatible reader, supporting selection, signature, email, phone, text, and numeric fields before or after payment. Developers configure forms with the Stripe API or Terminal SDKs, can sequence multiple inputs and attach metadata, and receive responses through terminal.reader.action\_succeeded webhooks; signature files are returned by file ID and need to be downloaded within 24 hours. The rental-car example uses selection and email forms for reservation lookup, followed by an on-reader agreement signature, replacing a separate signature pad and manual transcription while giving the POS more accurate data for receipts, lookups, and loyalty interactions.


### [Yes, You Can Use AI in Our Interviews. In fact, we insist](https://yomu.fyi/post/yes-you-can-use-ai-in-our-interviews-in-fact-we-insist.md)
- Company: [Canva](https://yomu.fyi/company/canva.md)
- Author: Simon Newton
- Published: Jun 11, 2025

Canva has updated its technical hiring process to require backend, machine learning, and frontend engineering candidates to utilize artificial intelligence tools such as Copilot, Cursor, and Claude during interviews. The transition addresses the limitations of traditional computer science fundamentals tests, which focused on writing algorithmic code from scratch even though AI assistants can generate complete solutions in seconds. To better evaluate on-the-job engineering capabilities, the company replaced its legacy screening with an AI-assisted coding competency that features complex, ambiguous product challenges such as designing an airport control system. Interviewers assess how candidates clarify requirements, guide tools on subtasks, debug flawed output, and verify that AI-generated code meets production quality standards. Candidates receive advance notice of these expectations, helping ensure prospective hires demonstrate strong engineering judgment when collaborating with assistive coding technologies.


### [Extending Docusign with Stripe to automate complex billing workflows](https://yomu.fyi/post/extending-docusign-with-stripe-to-automate-complex-billing-workflows.md)
- Company: [Stripe](https://yomu.fyi/company/stripe.md)
- Author: Paige Rossi
- Published: Jun 2, 2025

Docusign’s Stripe extension app addresses agreement workflows that continue beyond eSignature, where customer and billing data previously required manual transfer or complex integrations between platforms. The post shows how to install the app, start from the “Send new customer data to Stripe for invoicing” Maestro workflow template, and map web-form data to Stripe Customer, Invoice, InvoiceItem, and Email Invoice steps. It also explains how developers can switch the workflow to an API trigger, retrieve trigger requirements, and launch an instance with two authenticated API calls: one GET and one POST. The resulting flow collects customer information, creates a Stripe customer and draft invoice, adds an invoice item, and sends the invoice, while the returned instance URL can be opened or embedded for participants.


### [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

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.


### [How We Decomposed Tinder’s Monolith](https://yomu.fyi/post/how-we-decomposed-tinder-s-monolith.md)
- Company: [Tinder](https://yomu.fyi/company/tinder.md)
- Author: Tinder
- Published: May 23, 2025

Tinder faced significant agility and build performance challenges caused by an iOS codebase monolith containing over 1,000 files and 150,000 lines of code. Manual extraction efforts risked creating massive, unmanageable pull requests that would require constant rebasing against the main branch. To systematically decompose the target into Swift sub-modules, the team mapped declarations and references via the Swift compiler into a directed graph. They iteratively extracted leaf nodes with an in-degree of zero across sequential phases and automated common code adjustments, including module dependencies, imports, access control levels, and dependency injection. The automated decomposition completed in under six months with zero P0 incidents, reducing monolith build times by 78% and disallowing future additions to the monolith target.


### [Effortless enterprise authentication at Grab: Dex in action](https://yomu.fyi/post/effortless-enterprise-authentication-at-grab-dex-in-action.md)
- Company: [Grab](https://yomu.fyi/company/grab.md)
- Author: Kah Wei Lee
- Published: May 23, 2025

Grab needed a centralised system to simplify identity management, satisfy audit requirements, and standardise authentication across internal and external tools like Databricks and Datadog. The engineering team selected OpenID Connect as their standard protocol and adopted Dex, an open-source CNCF identity aggregator. Dex acts as an intermediary between applications and multiple identity providers to issue standardised OIDC tokens. To secure service-to-service communication, Grab implemented token exchange with trusted peer relationships rather than relying on privileged service accounts. Dex also provides a kill-switch mechanism that can route authentication traffic to an alternate provider during identity provider outages.


### [From failure to success: The birth of GrabGPT, Grab’s internal ChatGPT](https://yomu.fyi/post/from-failure-to-success-the-birth-of-grabgpt-grab-s-internal-chatgpt.md)
- Company: [Grab](https://yomu.fyi/company/grab.md)
- Author: Wenbo Wei
- Published: May 19, 2025

Grab's machine learning platform team initially faced overwhelming volumes of repetitive user inquiries across their internal support channels. An initial attempt to automate answers using the open-source chatbot-ui framework and GPT-3.5-turbo failed to scale because the 8,000-token context limit could not accommodate extensive documentation, and embedding search proved inadequate. The project then pivoted to create an internal conversational AI platform called GrabGPT by wiring chatbot-ui with Google authentication and Grab's catwalk model-serving infrastructure. The resulting internal service rapidly expanded across the organization, providing auditable interactions, multi-model support across OpenAI, Claude, and Gemini, and private network routing to safeguard corporate data.


[Newer posts](https://yomu.fyi/page/31.md) · [Older posts](https://yomu.fyi/page/33.md)