The Problem
Hybrid LSTM + LightGBM pipeline predicting Kubernetes pod failures before they happen, with Gemini-powered self-healing remediation.
AIOPS · ML SYSTEMS
KubePulse
KubePulse watches Kubernetes pod metrics in real time and predicts memory-related failures before they cause an outage. A hybrid LSTM + LightGBM model trained on Prometheus metrics flags anomalies ahead of the breach, a custom Prometheus exporter feeds the prediction back into the cluster for autoscaling, and Gemini 1.5 Pro generates the exact kubectl remediation command once an anomaly is confirmed.
GitHub · soonLive demo · soon
Stack
AI + ML
TensorFlowLightGBMLangChainGemini 1.5 Pro
Backend + Data
Python
Infrastructure
GoKubernetesPrometheusDocker
What I Built
KubePulse watches Kubernetes pod metrics in real time and predicts memory-related failures before they cause an outage. A hybrid LSTM + LightGBM model trained on Prometheus metrics…
The Difference
Hybrid LSTM + LightGBM model on Prometheus metrics cut CrashLoopBackOff events 38% versus threshold-based alerting
Numbers That Matter
38%primary scale
1.5measured result
40%system signal
E2Equality outcome
Engineering Notes ✎
Engineering Note 1
Hybrid LSTM + LightGBM model on Prometheus metrics cut CrashLoopBackOff events 38% versus threshold-based alerting
Engineering Note 2
Custom Prometheus exporter surfaces a predicted-memory metric that drives HPA autoscaling directly
Engineering Note 3
Gemini 1.5 Pro generates targeted kubectl remediation commands on confirmed anomalies, cutting MTTR 40%
Architecture Overview
Pod telemetry↓Anomaly Engine
LSTM ModelLightGBM ModelGemini Remediator
Prometheus↓Self-healing action