Juturna is still under heavy development, so there might be breaking changes along the way. We also recommend, for the moment, using Juturna in experimenting and prototyping scenarios. Please be sure you know what you are doing if you want to deploy Juturna in a production environment.

Juturna is a data pipeline library written in Python. It is particularly useful for fast prototyping multimedia, real-time data applications, as well as exploring and testing AI models, in a modular and flexible fashion.

Among its many features, there are a few keypoints to highligh about Juturna:

• ⚡ Real-Time Streaming: continuouusly process audio, video and arbitrary data streams
• 🔌 Modularity: create your own nodes and share them through the Juturna hub
• 🔗 Composable workloads: design pipelines to solve complex tasks in minutes
• 🚀 Parallelism & Batching: parallel, non-blocking execution for high throughput
• 📊 Observability: built-in logging and metrics support

documentation | contribute | meetecho

A pipeline can simply be defined as a collection of nodes.

Each node acquires a piece of data from its parent and, after performing a single task, provide its output to its children. In this sense, a Juturna pipeline is nothing else but a rooted tree, a particular kind of DAG where there is a single root node with in-degree of 0 (this is not technically the case, but we’ll skip it for now), and every other node has an in-degree of 1.

An example of one of our pipelines, currently in use for live audio transcription and summarisation is shown here.

Whilst Juturna ships with a number of built-in nodes (mainly source and sink nodes), you can implement your own nodes with ease, and share them so others can use them too.

To know more about the Juturna internals, please refer to the full documentation.

The following dependencies are required to use Juturna:

Juturna is currently available as a opensource codebase, but not yet published on PyPi. To install it on your system, first clone the repository, then use pip to install it (assuming you are working in a virtual environment):

(venv) $ git clone https://github.com/meetecho/juturna
(venv) $ pip install ./juturna

In case you want to include all the development dependencies in the installation, specify the dev group:

(venv) $ pip install "./juturna[dev]"

Alternatively, you can manually install the required dependencies, and just import the juturna module from within the repository folder:

(venv) $ pip install av ffmpeg-python opencv-python numpy requests websockets
(venv) $ python
>>> import juturna as jt

In the Juturna repository you will find a Dockerfile that can be used to create a base Juturna image. To build it, symply navigate within the repo folder and run:

$ docker build -t juturna:latest .

Please read CONTRIBUTING.md for:

• Branching & PR workflow
• Code style & linting (TBD)
• Issue triage (TBD)
• Issue & PR templates and a Code of Conduct are provided (TBD)
• Signing CRA

All notable changes are documented in CHANGELOG.md following Semantic Versioning.

Coming soon!

Distributed under the MIT License. See LICENSE for details.