Deploy-Log Root-Cause Agent

Debugging a failed deployment usually starts the same way: open the log, scroll through a wall of text, and manually piece together what went wrong. This project automates that first step. The Deploy-Log Root-Cause Agent takes any raw log — a Terraform apply failure, a Kubernetes pod crash, a GitHub Actions build error, an IAM permission denial — and returns a structured diagnosis: what failed, the likely root cause in plain language, a confidence level (high/medium/low) so it's honest about uncertain cases instead of guessing, a concrete actionable fix including a runnable CLI/Terraform/kubectl command where applicable, and a category tag. A lightweight retrieval layer checks the log against a small set of real past incidents before reasoning from scratch, using TF-IDF similarity. If the log resembles something seen before, the agent grounds its answer in that precedent and says so explicitly. If it doesn't, it reasons purely from the log itself rather than forcing a false match. The agent also detects which system produced the log (Terraform, Kubernetes, GitHub Actions, AWS, Azure, or GCP) and uses that as context for its reasoning. Every diagnosis can be rated useful or not, with feedback logged for future improvement, and can be automatically pushed to a Slack or Discord channel so a team sees failures the moment they're diagnosed, without anyone needing to go looking. Built with FastAPI, a Groq-hosted LLM (openai/gpt-oss-120b) via an OpenAI-compatible API with structured JSON output, scikit-learn for retrieval, and a lightweight vanilla JS frontend.