Skip to content

PRODUCT_DEMO_PLAN.md

Latest commit

 

History

History
107 lines (69 loc) · 3.51 KB

PRODUCT_DEMO_PLAN.md

File metadata and controls

107 lines (69 loc) · 3.51 KB
Raw

AiOps Product And Demo Plan

1) What this repo can already support

  • Simulate noisy workload behavior (network-noise, code-two/matrix).
  • Collect metrics with Prometheus/cAdvisor on a Kubernetes cluster running in Podman Desktop (kind-compatible).
  • Run a .NET orchestration sequence ("whirl") that triggers compute/agent actions (code-two/run/Program.cs).
  • Deploy baseline workloads to Kubernetes (deployment/*.yaml).

This is enough to build a credible "v1 customer demo" if we package it around one clear outcome.

2) Product definition (v1)

Product name (working)

  • bunnyhop aiops

One-line value proposition

  • Detect abnormal workload behavior and recommend (or apply) safer runtime settings before incidents escalate.

Ideal first customer

  • Platform/SRE teams running Kubernetes clusters with recurring CPU or latency incidents.
  • Teams without dedicated ML staff that still want practical optimization signals.

Narrow v1 use case

  • "Noise overload protection": detect sustained high CPU/noise behavior and trigger a controlled response recommendation.

3) Demo outcome to sell

Before

  • Workload shows unstable/high-noise utilization.

During

  • Metrics are ingested and analyzed.
  • Agent sequence produces a decision.

After

  • Customer sees either:
  1. A recommendation ("reduce loop intensity / adjust deployment values"), or
  2. An automated config change in a controlled namespace.

Keep the story around "faster mean time to mitigation" rather than "general AI."

4) 10-minute live demo structure

  1. Problem framing (1 min): recurring noisy workload incidents in K8s.
  2. Environment bring-up (2 min): start demo stack, show workload + metrics.
  3. Detection and analysis (3 min): run sequence, show metrics trend and agent state transition.
  4. Decision and action (2 min): show recommended or applied change.
  5. Business close (2 min): expected impact, integration path, next pilot milestone.

5) MVP backlog (customer-ready)

P0 (must-have)

  • Single command startup script for local demo.
  • Podman Desktop Kubernetes (kind-compatible) startup/teardown workflow with kubectl.
  • Stable deterministic demo data path (no surprises in live session).
  • One dashboard view (Grafana or CLI summary) with 3 KPIs:
    • workload CPU trend
    • anomaly window
    • mitigation decision
  • Decision log artifact saved to file (timestamp, signal, decision, confidence, action).
  • README rewrite focused on customer scenario, not internal notes.

P1 (high-value)

  • "recommend-only" and "auto-apply" modes.
  • Helm chart values patch generation from decision output.
  • Basic HTTP API for status and decision history.

P2 (later)

  • Multi-workload support.
  • Policy guardrails and approval workflow.
  • Tenant-aware dashboards.

6) Commercial packaging for customer conversations

Offer

  • 2-week pilot in one namespace, one workload class.

Success criteria

  • Detect and flag target anomaly class.
  • Produce actionable decision in less than 2 minutes from threshold breach.
  • Demonstrate at least one successful mitigation loop in controlled conditions.

Suggested pilot outputs

  • Pilot report (incident timeline + recommendations).
  • Operational runbook.
  • Rollout proposal for production guardrailed mode.

7) Immediate next steps (this repo)

  1. Create a deterministic demo script that starts stack, runs sequence, and captures outputs.
  2. Add a small decision log writer in code-two/run.
  3. Add a concise "Customer Demo Quickstart" to README.
  4. Rehearse a 10-minute script with strict timing.