Skip to main content
ARANO
← All tracks

Track · 10 modules

AI engineering for telecom

How to ship LLM and ML systems into the operator workflow without breaking things

For: RAN engineers learning ML/AI. Assumes deep telco knowledge; assumes Python literacy.

  1. 01

    Where LLMs belong in a RAN team

    The 5 tasks they're actually good at; the 3 tasks they're not.

    ~15 min○ planned
  2. 02

    RAG for network knowledge

    Embedding telco specs, hybrid retrieval, why cross-encoder re-ranking matters here.

    ~25 min○ planned
  3. 03

    Detector design — heuristic to ML to LLM

    When to use thresholds, when ML, when LLM reasoning. With three worked examples.

    ~25 min○ planned
  4. 04

    Eval-driven detector development

    How to grade detectors before shipping; model-graded vs golden-set evals.

    ~22 min○ planned
  5. 05

    Agents and orchestration

    Multi-agent vs single-LLM-with-tools, when each wins, how to wire them.

    ~22 min○ planned
  6. 06

    Cost and latency — the operator-grade view

    Local models on Ollama/vLLM vs cloud APIs, when each is the right answer.

    ~18 min○ planned
  7. 07

    Safety for autonomous network change

    Why "the model said so" isn't enough, the 6-layer safety stack pattern.

    ~25 min○ planned
  8. 08

    Observability for AI systems

    Langfuse, prompt regression catches, agent_run audit trails.

    ~18 min○ planned
  9. 09

    Data pipelines for PM/CM/FM at scale

    Parser, dedup, ROP-boundary alignment, vendor abstraction.

    ~25 min○ planned
  10. 10

    Going to production with operators

    Pilot scope, change windows, rollback, the human-in-the-loop reality.

    ~20 min○ planned