Michal Bartoň

Local storage

The local storage was used to store greatly reduced thumbnail versions of the images, which took up only a few tens of kilobytes. This made it possible to significantly reduce the data transfer between the user and the application as well as between the application and the cloud storage, especially in cases where the user only needed a quick preview without further interaction with the file.

At the same time, the local storage acted as a temporary cache for the processed files - for example, during translation, analysis, summarization or transcription. These files were automatically deleted when the task was completed.

3rd party APIs

Integration with third-party services was especially necessary when the tools were not open source or required specialized and powerful hardware to operate.

Language models from OpenAI and Anthropic have been used across most tools and workflows. For working with image content, the models from StabilityAI and Black Forest Labs were used - while these exist as open source, running them separately would require a significantly more demanding infrastructure than that provided by a regular web server.

Services such as DeepL were integrated for accurate translation, while Tavily Search was used to retrieve up-to-date and relevant information from the web.

FastAPI + Uvicorn/Gunicorn service

FastAPI was used for standalone instances of isolated tasks that often ran in the background - for example, text-to-speech (TTS), summarization, web scanning, or background removal from images. These jobs could be scaled independently and their requests were queued if they could not be processed immediately. This allowed computationally intensive operations to be handled safely and efficiently without risking overloading the main server.

Each microservice was deployed using the Gunicorn application proxy, which ran Uvicorn workers. Their number was configured individually according to the needs of the specific service. This architecture made it easy to move a given service to a separate server for better scaling if needed.

Cloud storage bucket + PostgreSQL database

The storage of generated, uploaded or edited files (e.g. images and audio recordings) and the database itself were placed in separate cloud instances. These services were only accessible to the application server for security reasons and allowed independent scaling as needed.

The repository was designed exclusively for final files intended for long-term preservation. Thumbnails, edits in progress, and generations in progress were temporarily cached until the user confirmed to save or rejected them. This minimized unnecessary data transfer and saved resources.

The database layer was served by Django ORM and managed by migrations. Data was encrypted and backed up regularly to ensure integrity and security.

Django backend

The backend of the application is built on the Django (Python) framework and forms the backbone of the entire application architecture. Its main responsibility is to securely mediate communication between the user, database, storage and external APIs.

It categorically provides data logic and persistence, user authentication and authorization, application security and protection, integration with external services, rendering and communication with the frontend, event and notification management.

HTML, CSS, JS frontend

The frontend included templates, which the backend used to render the base appearance of the page that the user viewed. It rendered the basic structure with the appearance. Sheets (datasets), modals (popups), workflows for tools were already managed/rendered by JavaScript, which only queried JSON data for the frontend.

The frontend logic was built on JavaScript, which provided dynamic data loading and thus relieved the backend from rendering large datasets. It also took care of displaying modals, managing workflow tools and overall interactivity of the interface. This made the application look and feel smooth and user-friendly.

UWSGI + NGINX

The application layer uses uWSGI as a powerful proxy server, configured for stable and scalable application operation.

The NGINX web server provides secure HTTPS access and efficient delivery of static files.

User

The system is designed to give users easy, secure and reliable access to all the tools and functions of the application. With a clear division of roles between backend, frontend and external services, the user can focus on the job at hand without worrying about technical details or security.