- Built an efficient and highly-scalable benchmarking infrastructure with Node.js. Kaskade uses Node’s worker threads for multi-threading, to allow one machine to simulate thousands of HTTP clients.- Implemented a graphic UI frontend with Electron + React. We use React to build a fast-responding and performant frontend. We use Electron so we can build our frontend as a desktop App while still leveraging the widely-used HTML / CSS / JS / React tech stack. This improves our development speed.- Our frontend streamlines the core user journey from configuring a benchmark to running and visualizing the result. It solves a major usability challenge of existing command-line based benchmarking tools.- Defined a modular benchmark config to allow users to reuse shared building blocks such as HTTP requests and sessions across multiple tests.- Designed SQL schema and use PostgreSQL to persist benchmark configs and historic results. Organizing benchmark results with a SQL database allows us to perform structural and complex queries over historic results.- To benchmark the performance of an HTTP server cluster, Kaskade can integrate with Kubernetes to run distributed tests from multiple machines and against multiple HTTP servers.

- Designed SQL database schemas to efficiently store, lookup, and update online shopping metadata, such as shopping carts, listings, inventory, and orders. We choose SQL since shopping data is mostly structural data that can be best organized, indexed, and queried with a relational data model. Moreover, ACID transaction guarantees are crucial in monetization procedures such as checkout and payment.- Integrated with Stripe to reuse a mature third-party online payment solution. Used a two-phase commit (2PC) protocol between POD and Stripe to properly handle failures during checkout and payment.- Used Amazon S3 Bucket to store and serve merchandise media such as photos and videos. This allows POD to store and serve a large amount of multimedia data with Cloud storage.- Built a multi-page web frontend with React + Redux, HTML, and CSS. Leveraged ReactRouter for client-side routing to achieve faster page navigation and a more streamlined user browsing experience.- Added third-party sign-up and authentication support with OAuth + Passport.js. POD users can log in with their commonly used web accounts, while POD doesn’t have to know those web accounts’ secrets.

- We designed Co-Sync to persist user data and to-do lists in MongoDB as separate documents. This is because to-do lists are not largely unstructured data. Storing them in a document store gives us more flexibility and better performance over a SQL database.- We designed Co-Sync as a single-page React web app. A single-page design provides a more uniform and immersive user experience. Once loaded, there is no more delay from page redirecting & reloading.- Since a single-page design needs to maintain more local states, we use Redux to manage frontend states and reduce development complexity.- We implemented the backend server with Node & express.js to serve HTTP requests and create, retrieve, update and delete To-do list items. Express.js’s middleware & routing feature allows us to better organize Co-Sync’s logic to put codes that serve related logic together. This reduces the cognitive cost and improves development velocity.

- Used PostgreSQL to transactionally store and update submissions’ states, comments, history, and other metadata. These are structural data that is best suited with SQL’s relational data model. ACID transactions guarantee correctness of critical actions such as committing a submission despite potential concurrent user actions or system failures.- Used React to build a multi-page web frontend. I chose React since it is easy to reuse common frontend components, such as submission list, file viewer, and comments box. Since frontend needs to maintain a large set and diverse state information, I used Redux to simplify the state management. - Implemented the backend server with Node and express.js framework. Using Node greatly simplified the development by just using one major programming language for both frontend & backend. This reduces the cognitive cost and simplified the codebase management.- DocRepo supports storing & organizing submitted documents on either local storage or cloud platforms (such as Google Cloud). This provides more options to DocRepo’s users to broaden its application scope, since different companies may have established different storage solutions.