RootCauseAI

The Problem I Wanted to Solve

When a production incident hits, L1/L2 support engineers waste hours jumping between systems. They check Jira for history, dig through git commits, query databases, read documentation—all manually. By the time they find the root cause, hours have passed.

Cloud-based AI tools could help, but many organisations (especially in regulated industries) can't send their code, logs, and database schemas to external APIs. They need something that runs entirely on-premise.

What I Built

RootCauseAI is a self-hosted investigation engine that:

• Runs locally using Ollama - No data leaves your infrastructure
• Connects to your existing tools - Jira, Redmine, Git, SQL databases, MongoDB
• Uses RAG for context - Indexes your documentation and past incidents
• Analyses code - Reviews relevant code paths for the failing component
• Suggests root causes - Based on patterns from similar past incidents

Technical Architecture

The key design decision was making it work entirely on-premise. Most businesses I've worked with in mining and resources can't send production data to cloud APIs. RootCauseAI runs as a Docker container using Ollama for inference, with connectors to common enterprise tools.

Languages supported: PHP, Java, C#, ASP.Net, Angular—the "boring" enterprise stack that most real businesses actually run.

What I Learned

Building RootCauseAI taught me that on-premise AI is viable for small teams, not just enterprises with GPU clusters. With quantised models running on Ollama, you can get useful AI assistance on a standard server. That's changed how I advise clients about data sovereignty options.