Agentic AI for 5G network root-cause analysis
My MSc research: a system that classifies live 5G network states from packet captures and uses retrieval-augmented agents to explain the root cause and suggest a fix.
- Context
- MSc Applied Artificial Intelligence dissertation, London South Bank University
- Year
- Year
- Methods
- ML classification, retrieval-augmented generation, agents
- Data
- Packet captures from testbed or dataset
network states classified: normal, congestion and flooding attack
classification accuracy on held-out data
agents explain the cause and propose mitigation
The question
When a network degrades, engineers have to work out what’s happening and why, often from raw packet data. Could a model classify the state of the network reliably, and could an agent then explain the cause in plain language and propose a mitigation?
What I built
A two-stage system. A machine learning classifier reads features from packet captures and labels the network state as normal operation, congestion or a flooding attack. A retrieval-augmented agent then takes that label and the evidence, retrieves relevant technical knowledge and writes an explanation of the likely root cause with suggested mitigations.
- Model and agent
- Engineer
- Network
Why it matters for clients
The same pattern works well beyond telecoms. A reliable model or rule set does the detection, then an agent explains the result and drafts the next step for a person to approve. It’s how I approach agentic systems in operations work.