AI-Powered Network Health Monitor

A three-stage Python pipeline. It pings the default gateway, public DNS resolvers and a live host to log latency and packet loss, runs an Isolation Forest per target to flag abnormal spikes without fixed thresholds, charts the results, and sends the summary to Llama 3.3 70B via Groq for a prioritised troubleshooting review.

Fixed latency thresholds are either too noisy or too quiet, and a chart on its own doesn't tell you what to do next.

Collect real ICMP data, let an unsupervised model decide what counts as abnormal for each target, and have an LLM turn the summary into a short, prioritised engineering review.

1_ping_collector.py

Pings each target at intervals and logs latency and packet loss to network_log.csv

2_analyze_anomalies.py

Isolation Forest per target, anomaly summary CSV and chart

3_senior_engineer_review.py

Sends the summary to Llama 3.3 70B via the Groq API for a written review

• Cross-platform subprocess handling for ping, with results logged to CSV.
• pandas for time-series processing; scikit-learn's IsolationForest adds an anomaly flag per row, per target.
• Matplotlib plots latency over time per target with anomalies highlighted.
• Prompt designed so the model responds like a senior network engineer: referencing the actual numbers, flagging risks and listing action items.
• The Groq API key is read from an environment variable; the repo ships a .env.example rather than a key.

• Collects latency and packet loss from gateway, Google and Cloudflare DNS, and a live host
• Unsupervised anomaly detection per target with Isolation Forest
• Latency-over-time chart with anomalies highlighted
• LLM review that references the data and lists prioritised actions