MessageCast and AgNet

4 min readNov 29, 2022

Enabling collaboration and machine learning in agriculture through open source and data syndication

Motivation

The global agriculture ecosystem creates over $5T in annual Gross Domestic Product (GDP), engages more than half the world’s labor population, consumes more than 70% of global fresh water and provides sustenance for the entire human population. This global ecosystem can benefit from digital solutions — powered by Artificial Intelligence and Machine Learning (AI/ML). The use of these applications can have a dramatically positive measurable impact on our water, food, and environment.

Challenge

There are two independent tool challenges:

A) There are no open source digital tools to set up an independent, customizable messaging app for agriculture-specific collaboration. For example, there aren’t any messaging tools to send or receive geospatial datatypes, tools that include a form-builder, and access to other datasets for agriculture-specific context.

From language translation to field-boundary tagging to user-discovery — such tools simply don’t exist in open source to enable anyone to quickly set up their own message server for agriculturally-rich context utilities — out of the box.

B) There is no syndicated, free tagged-data image repository for agriculture

There is a need for a tagged-data image repository with agriculture-specific context like plants, animals, diseases, and equipment AND with tagged content that can be leveraged to build agriculture applications easily and at scale.

While the tools are already available (e.g., PyTorch, TensorFlow, OpenAI), the data that is needed for this ML technology is unavailable.

Proposed Solution

Addressing the first challenge, AgStack is engaging its community of members, volunteers, and funding institutions around the world to co-create an agricultural context-rich open source messaging system — aka MessageCast. MessageCast is a downloadable tool that provides its partners and members (both in private, government and academic sectors), the ability to run a message server for their internal collaboration. Currently, the following features are included in this tool:

User discovery
Create topics and invite users to a topic
Create custom form data structures — e.g., alerts, forms, requests, surveys, etc. — all with no-code experience. Additionally, publish your custom forms in a forms library
Automatically encode location and geo-boundary (if available, the geo-asset-registry key) as part of the data. Permissively allow resolution to be 1 m to 1 km.
Access to documentation of usage including training videos and courses
Easy package management to download on a Linux environment (ubuntu 18.04 and later) — or easily spin a container instance on any participating public cloud vendor — and start using their own message server.

To address the other challenge, MessageCast automatically creates AgNet — an anonymous repository of each image tagged by the message content (no user attribution whatsoever). Images are added into an open data repository as a digital public good (DPG). This repository will be securely hosted by AgStack. Similar to ImageNet, AgNet will be available for use by any AI/ML application builder under the CDL 2.0 License created by the Linux Foundation. This repository will also have an API with integrations to all the public cloud providers who wish to register and co-create this.

The geocode for the image location will be embedded in the image metadata, and enable services like access to hourly weather data and access to soil attributes for that location.

Diagram of MessageCast and AgNet communication workflow

Making the content geo-tagged will enable accurate geo-context for plants, animals and equipment, along with access to weather, satellite, and soil information — all accessed with the GeoID of the field tagged in the metadata.

Vision: Start and scale

The key point here is that this DPG of a syndicated domain-centric tagged dataset requires collaboration between private, public, government, and other cloud providers. Syndication of this dataset through multiple distributed instances, that run independently of one another — while automatically sharing content in an anonymized gathering mechanism is one of the elegant ways. Based on several publications by McKinsey & Company and others — such applications can potentially unlock over $100B in value. Check out the Resources section for other articles and details.

We are currently inviting collaborators from public and private sector institutions and startups alike, at a global scale. We especially invite organizations participating in Agriculture Research and other Agriculture Extension, along with Cloud Native Computing Foundation (CNCF) ecosystems. We intend to leverage other Linux projects such as PyTorch/LFAI, Zephyr, LFEdge, Finos, OC-Climate, CNCF, Hyperledger, and others to bring the MessageCast vision to fruition.

Resources

Connect with us!

MessageCast and AgNet are necessary tools for the agriculture digital transformation. Be a part of the solution, join the conversation, and help develop MessageCast, AgNet, and other projects within AgStack. Receive more insight on how MessageCast and AgNet are being developed, and share your input. See the evolving blog, MessageCast and AgNet: Unleashing Collaboration, IoT and Machine Learning for Agriculture — Through Open Source.

Authors:

Sumer Johal: Executive Director at AgStack Foundation (sjohal@linuxfoundation.com),
Sandeep S. Joshi: CEO at AgPipe (sandeep@agpipe.com)
Mikela Jackson: Technical Writer at Red Hat (mzjackson@redhat.com)
Aanish Amir: Chief Maintainer at AgStack (aanishamir@gmail.com)