naveenck-10/recallops-ai

RecallOps AI is a prototype tool built for a hackathon that aims to help engineering teams stop solving the same production problems over and over. The idea is that when something breaks in a live system, engineers often already fixed a very similar issue months earlier, but that knowledge is scattered across chat messages, tickets, and people's memories instead of being searchable in one place. The tool works by letting someone paste in a raw error message or an alert from a monitoring system. It then searches a memory of past incidents using a semantic search library called Hindsight, which finds previous problems that are conceptually similar even if the wording is different, and feeds that context into an AI model. A separate routing library called CascadeFlow decides which underlying AI model should handle the request, sending simpler or already familiar problems to a faster and cheaper model while reserving a more powerful model for issues that look genuinely new, aiming to balance speed and cost. The AI models themselves are accessed through NVIDIA's API service. The result is a written explanation of the likely root cause along with a suggested step by step plan to resolve it, plus a dashboard showing how much AI usage, latency, and cost the system is consuming over time. The project is built with a React frontend and a Node.js and Express backend, and running it locally requires an NVIDIA API key along with separately starting both the backend and frontend servers. The author lists further planned work including direct integration with alerting tools like Slack and PagerDuty, and automatically generating infrastructure scripts to apply fixes, though none of that is built yet based on what is described in the README.