Articles
Worth reading
A short take on pieces worth your time — architecture, AI, system design, and engineering careers — each one linked back to its original source.
15 articles
Integrating Netflix's Foundation Model into Personalization Applications
Divya Gadde, Ko-Jen Hsiao, Dhaval Patel, and Moumita Bhattacharya walk through the three different ways Netflix integrated a single foundation model into multiple personalization surfaces, embeddings, subgraph, and fine-tuning, rather than training a bespoke model per use case. It's a practical answer to a question a lot of teams are facing right now: once you have one strong foundation model, how do you actually plug it into existing product surfaces without rebuilding everything downstream? The tradeoffs they lay out between the three integration patterns are transferable well beyond recommendation systems.
Netflix's Distributed Counter Abstraction
Netflix's engineers describe the counting service they built on top of their TimeSeries abstraction to handle 75,000 counter requests per second at single-digit-millisecond latency, the kind of "boring" infrastructure problem that turns out to be genuinely hard once you need it fast and consistent-enough at global scale. I like this as a case study in building a narrow, well-defined abstraction on top of an existing platform primitive instead of reinventing storage from scratch for every new use case. Relevant to anyone building shared counting or aggregation infrastructure.
Rebuilding messaging: How we designed our new system
Tyler Grant, Armen Hamstra, and Cliff Snyder open LinkedIn's messaging-rebuild series with the design phase: why a decade-old messaging system built for one use case couldn't cleanly extend to InMail, group chat, and integrations, and the architectural bets, a GraphQL API paired with a real-time stream, they made instead. What I find useful here is how much of the piece is about defining the problem precisely before touching the solution, a discipline that's easy to skip once a rewrite already has momentum behind it. Worth reading as a template for how to open a rewrite proposal, not just a rewrite postmortem.
Designing robust and predictable APIs with idempotency
The article that made idempotency keys a standard expectation for any payments-adjacent API: pass a client-generated key with a request, and retries — from a flaky network, a timeout, a naive retry loop — become safe instead of a double-charge waiting to happen. It's a deceptively simple pattern that solves a real and expensive class of production bugs, and this writeup is still one of the clearest explanations of why "just retry on failure" is dangerous without it. Essential if you're designing any API that touches money or triggers a side effect that can't be undone.
Online migrations at scale
Stripe's playbook for the unglamorous but high-stakes work of migrating a live, heavily-used data model — hundreds of millions of Subscriptions objects, no downtime, no data loss allowed — without ever pausing the business to do it. The dual-write-then-backfill-then-cutover pattern they describe is the one I reach for on every schema migration of consequence, because it's the only approach I've seen actually survive production edge cases. Read this before your next "just add a migration" turns into an incident.