Coursera

Right-Sizing AWS RDS? Request Mirroring Load Test Come To Rescue

Explains how to right-size AWS RDS instances and how to safely load-test different instance sizes by mirroring production traffic. Presents two practical mirroring approaches — Envoy (HTTP-level request mirroring to a clone app + clone DB) and ProxySQL (SQL-level mirroring to a clone DB) — and discusses metrics to watch, pros/cons, setup precautions (failover, double-writes, DNS/EC2 setup, extra network hops) and interpretation guidance.

Optimizing Low Code in Retool to Empower Services Colleagues

The post explains how Coursera’s Services team used the Retool low-code platform to build internal tools that abstract API and SQL complexity behind user-friendly interfaces. It details three concrete tools: a Learner Calendar that sequences six API calls from a single email input, a Translation Tagging tool that combines an ML similarity table with a JavaScript-built POST body to reduce errors, and a Degree Suppression List scripted to run daily suppression operations. The article notes on-premise deployment for data protection, use of Retool form triggers and conditional disabling to prevent user errors, and reliance on Retool documentation for custom scripting. It concludes with best practices focused on understanding database/API structure, designing the UI first, and preferring native Retool features before custom JavaScript.

Empowering Teammates and Thriving Remotely

The post describes practices for onboarding and collaborating in a remote-first engineering environment during an internship at Coursera. It highlights intentional one-on-one meetings, use of public Slack channels for questions, scheduled office hours, mentorship, and pair programming as mechanisms to share knowledge and unblock work. The author reports these practices improved rapport across teammates, increased cross-team learning, and helped them learn engineering problem-solving approaches. The piece discusses team and process techniques rather than technical architecture, algorithms, or code-level details.

Creating an Engineering Vision

Coursera describes its process for creating a three-year engineering vision and one-year engineering OKRs: differentiating vision vs. strategy, forming a core working group and area owners, assessing strengths/weaknesses via surveys and 1:1s, soliciting product roadmaps, using a "technology state of the union" one-pager, running async work followed by synchronous breakout sessions, gathering feedback, doing capacity planning, and iterating annually. Example outcomes include migrating from Cassandra to Aurora, improving test coverage, introducing SLOs, and building a component design system.

Analyzing the Impact of Course Updates with Difference-in-Differences

Coursera demonstrates using difference-in-differences (with regression and interaction terms) to estimate the effect of a major course update on monthly completion rate, verifying parallel trends with control courses and running the analysis in R or Python; they find an estimated 11% increase in completion.

Remote Internship Survival Guide

A Coursera intern describes challenges of remote work (isolation and blurred work–life boundaries) and offers practical tips: host virtual socials and one-on-ones to build team connections, use Zoom screen sharing for problem-solving, keep set working hours, separate workspace when possible, and take breaks to maintain balance.

Growing Coursera’s first team outside Silicon Valley

Get to know Coursera Engineer Bryan Rivera

The post is an interview with a Coursera front-end engineer who works on expanding and optimizing the company’s degree and MasterTrack products. He describes his role as a Front-End Software Engineer building comprehensive, accessible web pages for learners worldwide. He notes working in a shared codebase where mistakes can cause system failures and treats those failures as learning opportunities for debugging and improving code quality. He also gives practical advice on preparing for technical interviews, emphasizing timed practice and repeated exposure to build confidence.

Regression Discontinuity: Understanding the Benefit of Subtitles on Coursera

Coursera uses a regression discontinuity design around an 80% subtitled-video threshold (which triggers advertising the course as available in that language) to estimate the causal effect of subtitle availability on daily enrollments. They regress daily enrollments on percent subtitled, an indicator for above-80%, and their interaction to measure the jump at the threshold. The estimated effect is small; Coursera continues to promote subtitling via its Global Translator Community.

Improving Site Performance With Tree Shaking

Coursera migrated its front-end codebase to ES6 modules and used Webpack tree shaking plus related build-time techniques (Babel config, dynamic imports, babel-plugin-lodash, direct lodash module imports) to remove dead code and reduce bundle size. They describe migration tooling (codemod), a two-phased rollout to avoid breakage, edge cases they fixed, and measured results (up to 60% main bundle reduction for a high-traffic page and ~40% page speed improvement).

[Life@Coursera]: Why I chose Coursera as my first engineering job out of college

The author describes technical work from an internship that involved internationalizing the user experience for an admin platform. Engineers at Coursera emphasized ownership across the full delivery lifecycle, including development, QA, and deployment. The author is currently owning and piloting an experiment for a new search experience. The post notes mentorship and engineering conversations that informed team fit but contains no implementation details, code, architectures, or benchmarks.

Engineering Onboarding Bootcamp@Coursera

Coursera created a five-day, instructor-led engineering onboarding bootcamp (tracks for general/architecture, backend, frontend, and mobile) to reduce ramp time. The program runs monthly, is staffed by ~20 presenters delivering ~36 sessions, uses Google Sheets and calendar integration plus a workflow tool/script to automate scheduling, and supports remote delivery via Zoom. To date it has delivered ~150 hours of training to ~55 new engineers and interns.

Instrumental Variables & Randomized Encouragement Trials: Driving Engagement of Learners

Coursera used a randomized encouragement trial with an in-course message as an instrumental variable and two-stage least squares (2SLS) to test whether bingeing causes higher next-week completion; observational OLS showed positive correlation but the IV/2SLS estimate was not statistically significant, leading to the conclusion that Coursera should not actively encourage bingeing and instead support personalized schedules.

Order from Chaos: Understanding Search Queries through Vectors

The article describes Query2Vec, a pipeline that preprocesses Coursera search queries with NLP, generates vector representations using FastText (inspired by Word2Vec and trained using cosine-similarity-based, unsupervised approaches), aggregates phrase vectors, and uses Faiss to index and perform nearest-neighbor similarity search against a skills database. The embeddings are used to cluster and analyze query behavior to surface unmet learner demand and as features for ranking/recommendation/prediction models; the pipeline is run daily.

Evolving the Graph

Coursera describes a multi-year GraphQL journey: they initially built a thin, data-first GraphQL gateway that proxied many underlying REST APIs (using their Naptime framework) and generated schemas code-first with tooling like Sangria. They used namespace-based schema stitching and request-level federation (delegating to REST), and clients adopted Relay/Apollo. At scale the schema grew unwieldy (>7,000 types, >650 root types), exposing discoverability and governance issues. Coursera plans to move toward schema-driven (schema-first) development, stronger governance, and tooling changes to align GraphQL with product-centric workflows.

Improving End-to-End Testing at Coursera using Puppeteer and Jest

Coursera replaced a brittle Selenium/Sauce Labs e2e setup with a Puppeteer+Jest framework, added a local CLI and an AWS Lambda cloud test runner, integrated test runs into CI and code review (Phabricator), and built reporting/monitoring (dashboards, Slack notifications, retries) to reduce flakiness and improve developer experience.

An engineer’s perspective on engineering and data science collaboration for data products

An engineer’s perspective on effective collaboration between engineers and data scientists at Coursera when building data products. The post outlines three themes: define boundaries of focus (with empathy across roles), balance productionizing models with building platforms, and use data + SQL as the universal language. Examples include an experimentation platform, an automated coaching feature, and a recommendations module; the article also notes practical considerations like data warehouse democratization and who writes ETLs.

Two Summers at Coursera

Personal reflection by an intern at Coursera describing front-end work (React, Redux, Flow, GraphQL), investigation of a Slack integration, infrastructure work including a migration off Cassandra and use of an in-house system (Courral), developer-experience improvements around the Courier language and an IntelliJ plugin, and cultural elements like mentorship and Make-a-thon.

Improving The Learner Experience With Real-Time Subscription Renewals

Coursera replaced a polling-based subscription-renewal system with a webhook-driven, Kafka-backed pipeline so subscription status changes are handled in near real time. The new architecture reduces wasted PSP API calls, improves learner experience, and scales better; it includes a hybrid fallback poll job, progressive rollout, and extensive logging/metrics for observability and safety.