1. Introduction & The Problem

When production infrastructure fails, every minute of downtime costs thousands of dollars. While Site Reliability Engineers (SREs) write detailed postmortems after major incidents, these insights often get trapped in static wikis or forgotten Slack channels.

When a similar failure occurs months later, engineers start debugging from scratch — a phenomenon known as "Incident Amnesia".

For HackWithHyderabad 3.0 (Theme: "AI Agents That Learn Using Hindsight"), we built OpsMind (MemoraDesk) — an autonomous SRE AI agent that retains historical incident context, recalls past root causes during live outages, and reflects to create actionable prevention playbooks.

2. Architecture & Tech Stack

OpsMind connects an enterprise SRE dashboard to an agentic backend with persistent episodic memory:

  • Frontend: React 18, Vite, Tailwind CSS, Framer Motion, Lucide React.
  • Backend API: Python 3.12, FastAPI, SQLAlchemy async, Pydantic v2.
  • Memory Engine: Vectorize Hindsight Client (hindsight-client 1.2.0) utilizing retain, recall, and reflect.
  • Inference: Groq API (llama-3.3-70b-versatile) for rapid diagnostic generation.

3. How Hindsight Solves Incident Amnesia

Traditional RAG approaches only perform stateless similarity searches across documents. OpsMind leverages Hindsight's 3-phase memory model:

  1. Retain (aretain): When an incident is investigated and resolved, OpsMind commits the incident telemetry, error logs, and mitigation steps into Hindsight episodic memory banks.
  2. Recall (arecall): When a new P1/P2 outage triggers an alert, OpsMind automatically recalls past matching root causes, related runbooks, and previous resolution strategies.
  3. Reflect (areflect): After postmortems are reviewed, OpsMind triggers background reflection jobs that synthesize recurring failure patterns into preventive runbooks.

4. Key Results & Impact

  • Reduced MTTR (Mean Time to Resolution): On-call engineers instantly receive past resolution context instead of manually searching old wikis.
  • Continuous Learning: The more incidents OpsMind investigates, the smarter and more accurate its diagnostics become.

5. Repository Links