Software Developer in Brisbane, Queensland, Australia

I built a customized static website generator that received the federated GraphQL schema and produced a responsive static website containing the API documentation. I developed the site-building tool and the front-end UI. Project milestones varied from accessing the federated registry directly to moving from legacy Mustache.js templates to server-side React. I created a complete set of tests from scratch. These ranged from unit tests parsing the GraphQL schema to snapshot tests comparing site pages from a static schema.

An RSS reader for browsing the news. I have used versions of it personally as a replacement for Google Reader since that service was discontinued in 2014. I am the only person involved in the project.

A current back end uses Cloudflare's serverless workers to provide a GraphQL API. All state is kept in a key value store. The total response time to a client is around 100ms for queries, while mutations that require polling RSS feeds are bottlenecked by the slowest feed response.

Front ends have been made in React and Next.js, React Native, and Swift UI. In each implementation, the UI is optimistic, with Apollo's client cache reflecting user actions across the entire app immediately. iOS implementations have features like pull-to-refetch latest articles on the 'News' page and slide-right to delete a subscribed feed on the 'Subscriptions' page.

The back end can be browsed at https://worker-rss.mikowals.workers.dev/__graphql with the code repository at https://github.com/mikowals/cloudflare-worker-rss.

The React Native front end for iOS is at exp://exp.host/@mikowals/rss-react-native.

The code repository for the React and Next.js is at https://github.com/mikowals/simple-rss-next. It is not currently running anywhere.

The application was a locality-based social network that allowed people to connect and plan events. I joined the project already in progress and implemented features such as iOS notifications using Firebase and completing the build process to get the app into TestFlight for testing.

I also did several rounds of bug fixes for the app. For example, the chat features would lose messages, so I debugged the Websocket connection implemented in socket.io to eliminate the problem.

The app will be released in the App Store shortly.

An iOS and Android game built to help expecting parents choose a name. Along with a partner, I designed, developed, and marketed the game from concept to launch. The game was available for download for several years. It had hundreds of people playing daily and attracted 100,000+ downloads.

Requirements:
• MongoDB database to host names, customer data, and game state.
• A single-page application wrapped for iOS and Android store downloads.
• Hosting with release and version migration scripts on the Meteor platform and DigitalOcean.

Game features:
• Freemium model progressing from anonymous play to paid customer.
• Joint play with a secure invite system shared data between linked players.
• Responsive UI for multiple phone and tablet sizes.
• Fast game play with optimistic server interactions and user milestone updates.
• Analytics and reporting interface to track usage and progress through an upgrade cycle.
• Available in multiple countries with geographically appropriate names.

A package for the Meteor JavaScript framework that makes inserting a large number of documents into MongoDB more efficient (100x improvement). The package maintains Meteor database security rules and enables Meteors' optimistic UI features for batch inserts from the client.

The package has been used by 5,000+ developers and has been maintained since 2015.

The project asked whether carefully allocating a bond's duration risk across multiple buckets would improve hedging efficiency. Hedge efficiency means how accurately the price change of a portfolio matches the estimated change based on duration and yield change.

Working in C and C++, I built a back-testing harness and point-in-time model to produce the historic ratios for trades hedged with standard and bucket durations. Backtesting could report the daily standard deviation of realized profit from expected profit.

The project discovered that the firm's treasury curve model had a computational quirk where minor price changes could cause the curve to "pop" between curling up and curling down. The effect of this quirk on hedge efficiency was three times larger than the duration effects originally studied.

This inefficiency was fixed by adding an extra step to the treasury model that forced the curve to pass through the on-the-run (most liquid) bond yields.

An iOS app that helps users tracks fitness goals. The app was available for a number of years in several countries through the App Store. A nice technical feature of the app was its animated page transitions while users were looking through graphs of their progress.

I contributed pull requests and bug reports to TensorFlow, Swift for TensorFlow (S4TF), PyTorch, and other academic machine learning projects.

My contributions focused on implementing high-level machine learning algorithms, but given the newness of the software, this often led to fixing framework bugs. For example, I had pull requests accepted for TensorFlow and Swift for TensorFlow to fix their respective implementations of batch normalization. The fix in TensorFlow corrected the application of different averages for training versus inference, while the fix in S4TF corrected the automatic derivative calculation for a square root.

The client was a fintech company with several successful services that wanted to launch a new mobile app for a new service idea. The deliverables were:

1) A business plan outlining competitors approaches, market segments, and a business model;
2) A specification for the app with front-end designs in Figma, a database schema, and back-end API documentation in OpenAPI;
3) Initial code for the app with TypeScript/React Native, AWS back-end APIs, and AWS CloudFormation scripts to run CI/CD.

A game involving decisions about faces built to gather research data for a large team of academics in economics, psychology, and machine learning. The deliverables consisted of:

• React responsive web app allowing basic profile management and fast gameplay.
• Node.js back-end with WebSocket data synchronization to the client and a REST API to connect with an AI player microservice.
• Automated end-to-end tests using Playwright. Automated unit and integration tests using Mocha and Sinon.JS.
• GitLab CI/CD enables multiple separate instances of the site to run for data gathering and feature testing.
• MongoDB Atlas charts to visualize collected data.