Pascal Heitz

• Introduction of TypeScript and NestJS into the backend ecosystem (initially legacy JavaScript / Express).
• Mentoring of long-standing developers on new practices and technologies.
• Rewriting and rationalization of legacy microservices.
• Migration of internal CommonJS libraries to TypeScript libraries that are future-ready while remaining backward-compatible with legacy CJS deployments.
• Creation of an extended library of recurring features for NestJS applications, including:
  • A NestJS application creation wrapper for AWS Lambda, with standardized multi-input handling (HTTP, SQS, SNS, Cron) and numerous modular plugin-based features.
  • A reusable internal ESLint package.
  • NestJS services for SNS / SQS (AWS) event multiplexing and demultiplexing.
  • Strongly typed, data-safe NestJS services encapsulating legacy JavaScript components.
  • Reusable guards and interceptors.
• Creation of a library of composable mocks, simplifying the construction of unit test scenarios.
• Implementation of an abstraction layer on top of Sequelize, enabling automatic inference of TypeScript types for queries, including recursive includes.

Users get quized on their knowledge of important facts of League of legends.

Supermemo is used to keep track of the user's progression on the toughest questions.

On the backend side, a react-admin admin panel allows admins to create and edit questions.

Demo: Lol Trainer

I have been learning Czech for the last couple of years. Watching Czech movies has been very tedious, as I was encountering an overwhelming number of unknown words every minute.

There was no good tool to assist me creating Anki flashcard. I've tried every tool I could find, but there were mostly buggy and abandoned projects. So I decided to build my own.

Under the hood, it uses LibVLC and Chromium glued with WPF's bindings.

Give it a go at flashyplayer.com.

Ordamo, a London-based chain of restaurants makes extensive use of Android tablets to manage orders. These hundreds of tablets were managed with a commercial MDM (Mobile Device Management) solution. This MDM bills their service on a per-device-per-month basis and this makes it a very expensive setup.

Ultimately we needed to build our in-house MDM solution for multiple benefits:

• Replace recurrent high running costs by a fixed development cost.
• Create features that the current commercial MDM solution does not provide.

Challenges were numerous:

• Multiple ways to achieve tablet management: Android Enterprise, G-Suite, Android Management API, Administrator app, device owner app, etc. Ultimately, we decided to use the Device Owner app, which combines powerful features, and no running costs.
• Each device and version of android have their quirks. Some features are only available in late versions. Some available in early versions were removed in later versions (mainly for safety and privacy). Many workarounds were necessary to ensure the same features on every device and Android version.
• The Android documentation is very scarce when it comes to administration features. Many trial-and-errors were required to empirically figure out undocumented behaviors.

In addition to the MDM solution, a standalone Android app was made to quickly enroll new devices with either NFC or a QR-Code. This gave the opportunity to try Ionic 5 and Capacitor which were just released.

"Pick the perfect wine, every time"

Corkscrew helps you pick the best available wine for any food you choose in a restaurant.

To do so, it regularly fetches the latest wine lists in the restaurants' websites.

Among the small Corkscrew team, my responsibilities were multiple:

• Create, maintain and update the app's backend and the administration dashboard. Design efficient processes to curate incoming data. Angular.js, Node.js, MongoDB.
• Keep the database's wine data consistent and normalized. Fix data conflicts and reality / model mapping. Build automatic import tools to enhance the automatic wine recognition.
• Create Guzzler, website crawler regularly fetching wines from e-shops.
• Build a rule-based algorithm to match a wine name on a wine list against a wine in our database. Our database contains thousands of named entities. It also understands wine list layouts, glass / bottle volumes, prices, etc.

Give it a go at Corkscrew.app.

Took part in the development of the in-house Brain Simulator.

• CUDA implementation of convolution neural networks and autoencoders.
• Live visualization of the network's state (e.g. weights, errors).
• R&D in reinforcement learning around DeepMind's paper.

Speed cameras' aperture and gain are controlled by a physical lux meter. The lux meter is an expensive piece of hardware to buy, setup and maintain. The goal was to determine an appropriate method to choose the best aperture and gain according to the images coming from the camera itself, without the need for any external hardware.