Ivan Zderadicka

This is a solution for automatically checking PDF docs. Users can upload a file and numerous custom checks are performed (like searching for unwanted words or missing words, checking date and number formats consistency, checking price calculations, and so on).

PDF Checker is written in Python, and the key component is a PDF parser, which converts PDF doc into text (with additional metadata about text font and position on the page). Then there is a flexible system of plugins, which implement individual checks and each plugin receives the parsed text and can apply its logic. All available plugins/checks are then executed for a given doc resulting in a user report.

This provides an easy-to-use web interface (HTML5 and jQuery on the client-side, Python and Flask on the server-side). This enables users to upload a document and examine found problems—problems are highlighted directly on the original PDF doc with appropriate comments.

It is also performant; it can make tens of checks in seconds on 100-page docs.

As part of this project, I built a CI/CD infrastructure (based on XenServer, GitLab, and Docker), which enabled us to implement and deploy new checks very quickly (in a couple of hours).

Mybookshelf2 is a platform for the management and sharing of eBooks. It consists of several parts: server (Python, Flask, SQLAlchemy, PostgreSQL), long-running tasks scheduler (Python, Asyncio, WebSockets; it's basically my separate project, Asexor) and SPA web client (HTML5, CSS3, Bootstrap 3, Aurelia framework).

Mybookshelf2 provides easy eBook browsing (by authors, series, genres, and so on) and a full-text search.

Users can create and share their own lists (so-called bookshelves). Special advantages of Mybookshelf2 are advanced eBook file upload, where all of the information is extracted from file metadata, so an ebook can be uploaded with just a few clicks. You can also merge eBooks and authors together to maintain collection consistency.

Mybookshelf2 is unique among similar solutions because it enables the user to maintain several files of the same type for one eBook and also enables interactive and batch transformations between eBook file formats.

Audioserve is an open-source solution for streaming audiobooks. The server was written in Rust using asynchronous IO like Tokio and Hyper libraries and provides a highly performant network interface.

Audioserve can transcode audio files as part of the streaming process, thus significantly saving bandwidth while maintaining high audio quality. The Rust server is very efficient and only takes a minimum of system resources.

Audioserve provides a light web client using jQuery, Bootstrap 4, and Android client written in Kotlin. The Android client implements smart caching, enabling users to listen to audiobooks without interruption, even when the connection drops.

imap_detach is a universal tool for batch email attachment downloads. Compared to similar tools, it provides advanced attachment filtering using IMAP filters (enabling that only desired message parts are downloaded from the server) and concurrent downloading of multiple attachments.

It implements simple-to-use logical filters expressions, but users can use them to create quite complex conditions for required attachments.

The tool was implemented in Python.

adecaptcha is a toolset for experimenting with automated audio captcha recognition. It's written in Python, NumPy, SciPy, and PyQt.

It extracts audio signal features using MFCCs (mel-frequency cepstral coefficients) method and classifies them with the SVM (support vector machine) algorithm.

The package contains libraries for signal processing and classification and the GUI tool for the preparation of training data sets.

In my free time, I've contributed to open-source which helped keep my technical skills up to date and hopefully also provide some useful stuff back to the community. See my GitHub profile for details.

• https://github.com/izderadicka

I helped migrate from monolithic IOT Python application into componentized architecture running in Kubernetes containers. The final architecture used IoT core services of Google Cloud, seamlessly connected to TensorFlow data processing and simple-yet-efficient measurements preprocessing and batching implemented in fully configurable Kubernetes services.

The key benefits were improved processing time for each IoT measurement event with a speedup of about 3x and enhanced reliability and monitoring using Google Stackdriver.