Provide a simple way of ML and network engineers to collaborate on AI-driven network management.

This platform brings AI-driven analytics and automatic decision creation to telecommunications networks.

Network telemetry is continuously ingested, processed, and stored. Machine learning models run inference over that data and publish results to a shared event stream. A policy layer governs data access between all components, ensuring that only authorized services can read or act on sensitive network metrics. An LLM-based decision service consumes model inferences and translates them into actionable network recommendations.

The result is a closed-loop system where raw network data flows through collection, analysis, and decision stages, with policy enforcement and observability at every step.

For a more detailed explanation of the project access our website.

• Docker compose
• git
• make (optional)
• CPU: 4+ cores
• RAM: 16 GB recommended
• Disk: 20+ GB free
• Ollama instance ( Just needed for decisions )

git clone --recurse-submodules https://github.com/ATNoG/pei-nwdaf

Note: The .env.example can be used as .env for single-machine deployments. All services communicate via the internal nwdaf-network Docker bridge network.

make env 
# edit .env as needed

make dev

All services expose ports for debugging. Access services directly on their configured ports.

make prod 

Note that in this mode we only expose ports for nginx, ingestion service , decision service and observability stack.

As for now we only support Nef_event_exposure on ingestion service. (without auth). If you don't have one then you can run the simulator we provide. Note that this simulator follows TS 29.591 but only sends random values.

make producers

Go to http://localhost/

Note: To fully understand the project, we highly recommend reading the microsite 🙂.

See CONTRIBUTING.md.

License