Why I'm Building My Own GPT-based Troubleshooting Agent

Why I'm Building My Own GPT-based Troubleshooting Agent


RAG GPT DevOps Pinecone

Troubleshooting complex systems has always been a major part of my job—whether during my years at Amazon Web Service(AWS) helping global enterprises resolve urgent technical issues, or now at National Broadband Ireland, supporting national infrastructure in a mission-critical DevOps role.

But the tools we rely on—manual document searches, outdated Confluence pages, scattered logs—just don’t scale with the complexity of modern distributed systems.

That’s why I built my own GPT-based troubleshooting agent: a fully personalized AI system capable of understanding, remembering, and evolving with the way I work.

The system—CMAI GPT—ingests my documents, high-level designs, logs, and even pipeline specs, transforming them into searchable vector embeddings with Pinecone. It then uses OpenAI’s GPT-4 Turbo for retrieval-augmented generation (RAG), delivering precise, context-aware answers to my queries in real-time.

But I didn’t stop there.

I added:

  • Lambda-based embedding pipelines
  • FastAPI-powered endpoints
  • Semantic memory that learns from feedback
  • Dynamic document ingestion and auto-indexing
  • Vector drift awareness and self-reinforcement loops

This isn’t just a knowledge base—it’s a self-evolving AI partner that grows smarter as I work. It can explain edge-case bugs, trace architectural dependencies, and answer questions faster than any team wiki ever could.

The deeper I built, the clearer my goal became: I want to develop AI agents that enhance human cognition. Tools that help engineers like me think more clearly, solve faster, and retain institutional knowledge over time.

This project is also what inspired me to pursue formal AI education. I want to strengthen my theoretical grounding—not just in GPTs and embeddings, but in model architectures, RLHF, and responsible deployment strategies.

Because I believe we are at the edge of something transformative—not just for productivity, but for how we transfer knowledge, accelerate expertise, and empower individuals and nations alike.

This agent is my first step.

© 2025 Charlie Mai
Application portfolio for AI-focused MSc programmes at Trinity College Dublin, University College Dublin, Dublin City University, , University of Limerick, University of Galway.
Email: me@cmai.ai