← All projects

AIOps — anomaly detection and an LLM runbook assistant

In build — target Sept 13–14, 2026. The plan below is complete; the working detector, assistant transcripts and screenshots land here as the build finishes.

Python anomaly detection LLM API RAG

What this project builds

Two parts. Part A — unsupervised anomaly detection over log-derived metrics, running locally: extract metrics from operational logs, train a compact model, flag the anomalies, plot the results. Part B — a retrieval-augmented assistant that answers "what do I do about this alert?" from my own runbooks, with citations.

How it will be built

  • Part A in plain Python: generate_sample_logs.pydetect_anomalies.pyplot_results.py; the detection model itself is a handful of lines — the craft is in the feature extraction.
  • Part B: my real runbooks embedded and retrieved, an LLM API composing grounded answers that cite the runbook they came from. Under USD 5 total in API calls.

Why it matters

Most AIOps demos have to invent their domain. I have the ingredient that can't be faked: years of real operational logs and real runbooks from production systems. This project connects that operational depth to the AI tooling wave — the combination the market is starting to hire for.