Juan Carlos Jiménez
← All work

01 / 11 · INCIDENTS

Incident detection

An LLM pipeline that reads real customer cases and raises incidents before they spread.

Context
Telefónica / Movistar Plus
Role
Design and production
Year
2024
Stack
Azure OpenAI · Python · Azure Functions · REST APIs

The problem

Incidents surfaced late. By the time a pattern was obvious in the case queue, many customers had already called.

What I did

I designed the detection flow on Azure OpenAI over incoming cases, wrote and evaluated the prompts against labelled history, and shipped it as a service the operations team checks every day.

What changed

It runs in production and its alerts are trusted enough to act on. The team investigates what it raises instead of hunting for it.