Platform Alpha

Showcase Application

SingularityNET Integration

NuNet Platform Private Alpha

Dynamic Service Deployment

NuNet uses distributed AI services hosted on SingularityNET as docker containers, and deploys them on a decentralized network of computers with diverse hardware. The Private Alpha is a demo of a full serverless computing solution that avoids lock-in to any one vendor, and promotes seamless interoperation between machines owned and operated by different owners on an open internet. Further versions of the platform will address end-to-end security of each workflow and sandboxed computing process.

See video and its slides for an in-depth understanding of NuNet’s service discovery principles.

Workflow construction

The Private Alpha demonstrates how complex application logic and computational workflows – involving many AI services on many different machines – can be constructed by simply declaring a program graph and letting the platform resolve the dependencies. NuNet will automatically scale this hardware/software service mesh to the dynamic requirements of the application. The Fake News Warning application showcases a browser-extension with a distributed workflow being dynamically constructed and deployed on the network.

Token compensation for service providers

The Private Alpha implements and demonstrates the basics of internal tokenomics and supply-demand dynamics of the platform. Machines can automatically enter into a smart contract which holds payment in escrow for each workflow, and releases it to each machine in proportion to the work it contributed, provided that correctly formatted results are returned. The Alpha showcases a smart contract that mines NuNet test tokens and sends them to machine owners who complete work. In the next versions of the platform, we will add a generic gateway which will allow any SingularityNET AI services (and later services on other frameworks) to automatically send a portion of the payment they receive to cover their computing costs.

Network status monitoring and process tracking

All activity on NuNet is monitored by publicly-available network status dashboard. With some resemblance to blockchain monitoring services, the dashboard registers each computational process carried out by the machines on NuNet, the precise telemetry and costs of each process, overall network utilization statistics and computing power. Alpha utilizes the network status dashboard that was developed by NuNet for its first Minimal Viable Product. In future releases, the platform will feature individual dashboards to allow each user (consumers and providers) to explore full telemetry, earnings and costs of operations.

New device onboarding procedure

Any entity, including individuals, small or large businesses and organizations, will be able to contribute a precisely defined amount of their computing resources to NuNet’s network by running a script that performs the required benchmarking, asks and records owner preferences, and permitting NuNet to deploy fitting AI services with these resources. With private Alpha we are releasing a demo script for Linux boxes.

Point-to-point communication between machines

A crucial aspect of NuNet platform is that it allows privately-owned machines behind Network Address Translation (NAT) routers to join the network and discover and be discovered by other private devices. The backbone of NuNet’s technology is secure P2P communication between public and private machines in the network. The private Alpha enables adapters running on each separate machine with private or public Internet address to communicate over an overlay network.

Tooling for developers and contributors

As a platform that connects entrepreneurs, developers and machine owners, NuNet is not aiming to develop concrete use-cases, workflows and programs, but rather to provide all participants on the platform with adequate resources to achieve their purposes. With Alpha, we developed the following features (and used them ourselves to build the showcase application):

Private testnet

We run a private Ethereum network for developing, debugging and monitoring the smart contracts of each workflow, and to benchmark and adjust the tokenomic dynamics of the whole network. Post-private-Alpha versions of the platform will ship the developed tokenomic system to public blockchains, but will continue to maintain and provide access to multiple private testchains (whether Ethereum or other blockchains) as a part of developer tooling.

Full access to message traces and service logs

Applications running on NuNet will be truly decentralized, forming a dynamic hardware/software mesh with multiple endpoints and geographically dispersed topologies. For developing, debugging and monitoring such distributed applications, NuNet has built infrastructure and tools for distributed tracing and logging, based on the industry standard OpenTelemetry standard and Jaeger distributed tracing framework. These tools are implemented in the actual test call trace of the fake news warning app.

Full contributor access to development process

Last but not least, NuNet is opening developer access to its platform development repositories and git process. Application developers and technically-oriented community members will have full access to the open source code, issue and ticket tracking systems and internal technical discussions.

Fake news warning application

NuNet has developed a showcase application which is included in the pre-Alpha release – the fake news warning app. This app demonstrates the platform’s ability to dynamically construct application-specific service meshes from several SingularityNET AI services. As introduced in the description of the private Alpha, in order for the service mesh to be deployed, developers have to:

  1. Describe the application in terms of a program graph – defining AI Services and program components;
  2. Make sure that all AI Services in the program graph are onboarded onto SingularityNET;
  3. Build a frontend of the application that exposes its functionality to users.

The fake news warning application showcases these three aspects. The frontend is a browser extension that, upon activation, sends the contents of the current browser tab to the currently pre-defined program graph, consisting of open-source algorithms. These algorithms are onboarded onto SingularityNET, and run on NuNet’s test network. Each AI service of the fake news warning app evaluates the probability that provided content contains fake news and employs a different theoretical and computational approach for that. Currently, the fake news warning app combines a stance detection algorithm with neural-network based binary classification to come up with an estimation. This architecture is illustrated in the figure below:

A program graph that constructs this architecture is:

"dag": {
  "news-score" : ["uclnlp","binary-classification"]

Importantly, the fake news warning application’s backend can adapt and evolve. It is constructed dynamically from interactions between SingularityNET and NuNet, with SingularityNET providing AI algorithms and NuNet deploying them on decentralized hardware, so it can be easily reconfigured by declaring a different program graph. The private Alpha will construct the backend dynamically from the declared program graph. Furthermore, the browser extension component of the app will list all returned results from the backend, and related analysis of the call according to an augmented program graph. A user will be able to see complete raw data of each call and its return and provide granular feedback:

The current precision of the fake news warning algorithm, developed by the NuNet team, is less important to the demo than the functionality of the Alpha platform itself. The platform we are demonstrating allows a community of developers to offer more and better fake-news-warning algorithms, and integrate them into the application that is readily available for users. Users of the app can select the most fitting algorithms by providing their feedback and paying service calls.

Download and install the latest version of fake news warning browser extension from here.

The fake news warning app showcases a general pattern of application development that NuNet adds to the SingularityNET ecosystem. Obviously this is only possible by closely integrating NuNet’s platform and tokenomics with SingularityNET.

Integration to SingularityNET ecosystem

Our mission is to integrate the world’s computing resources, data, and ingenuity into a single network tied together by a powerful API of APIs. We first aim to bring these resources into the SingularityNET ecosystem. The main aspects of NuNet and SingularityNET’s integration are:

  • SingularityNET AI Services will be able to issue calls to NuNet in order to run on decentralized hardware;
  • Developers in the SingularityNET community will be able to propose their algorithms use NuNet for serverless deployment or to deploy their services themselves;
  • NuNet will adopt, integrate and participate in the further development of AI-DSL – a major research and development avenue of SingularityNET – in order to enable decentralized hardware and data resources to collaborate with AI services in an autonomous way;

NuNet enriches SingularityNET’s ecosystem and its possibilities. Together, SingularityNET and NuNet provide software and hardware to the community to build versatile and efficient AI applications and keep them decentralized. The NuNet Alpha, and the fake news warning application built on it, serve as proof of concept of this broad potential.