$whoami
← All projects
AI · Cloud Cost Optimizationconceptualized

FinOps Intelligence Agent

AI-assisted cloud waste detection, spend anomaly analysis, and actionable savings recommendations.

View on GitHub →Discuss this project
$cat problem.md

Problem

Cloud spend grew faster than usage visibility. Idle EBS volumes, oversized instances, and orphaned resources accumulated without proactive detection.

$cat solution.md

Solution

Designed an AI agent that ingests billing and utilization metrics, correlates usage patterns, flags waste and anomalies, and generates remediation runbooks with estimated savings.

$cat how-it-works.md

How it works

1

Daily scan

Step Functions orchestrates resource metric collection across accounts and regions.

2

AI analysis

Agent correlates utilization vs billing, flags idle resources and spend anomalies.

3

Actionable report

Savings report with remediation runbooks pushed to Slack and Jira for team review.

System architecture

Cloud APIFinOps AIDetectReportSavings

Cloud APIs (Cost Explorer, CloudWatch) → scheduled ingestion → FinOps AI agent → waste detection + anomaly scoring → savings report + remediation workflow.

Operations dashboard

finops-agent · cost dashboard

Est. savings

$12K

Flagged

47

Anomalies

3

Top waste categories

Idle EBS · $420/mo

Oversized EC2 · $280/mo

Orphaned EIP · $45/mo

Operational dashboard — request queue, agent status, and audit trail
Deployment flow

01

Schedule

EventBridge cron

02

Ingest

Cost Explorer API

03

Agent

Correlate + score

04

Report

JSON + dashboard

05

Alert

Slack/Jira

Implementation details

  • Step Functions state machine for multi-account scan orchestration
  • Baseline learning per account stored in S3 with 30-day rolling windows
  • Detection rules versioned separately from agent prompt for testability
  • Closed-loop tracking: estimated vs realized savings per remediation

Proof of work

finops-agent.log
$ finops-agent scan --account prod-123456
→ scanning: 1,842 resources across 3 regions
→ idle: 3 EBS volumes (450GB) · est. $420/mo
→ idle: 2 t3.xlarge (avg CPU 2%) · est. $280/mo
→ anomaly: S3 egress +340% vs 30d baseline
→ report: finops-rpt-2026-05-28.json
status: COMPLETE · actionable items: 12
FinOps agent scan output — idle resources and spend anomalies with savings estimates.

Workflow

  1. 1

    Scheduled scan

    Pull resource metrics and billing data from AWS Cost Explorer and CloudWatch on a daily cadence.

  2. 2

    AI correlation

    Agent correlates utilization patterns, flags idle resources, and scores spend anomalies against baselines.

  3. 3

    Report & remediate

    Generate savings report with estimated impact and link to automated or manual remediation runbooks.

Observability integration

  • Scan completion metrics and resource coverage percentage
  • False-positive rate tracking per detection rule
  • Savings realized vs estimated (closed-loop FinOps)

Design decisions

  • Separate detection from remediation — reports first, actions gated by policy
  • Baseline learning per account/team to reduce false positives
  • Savings estimates tied to actual billing line items, not generic heuristics
  • Integration hooks for Jira/Slack alerting on threshold breaches

Inputs

Cloud billing dataResource metricsUsage thresholdsTeam ownership tags

Outputs

Waste reportSavings estimatesRemediation runbooksAlert events

Tech stack

PythonAWS Cost ExplorerCloudWatchClaudeStep Functions

Impact

$12K/mo

Estimated idle resource savings

47

Resources flagged in pilot scan

3

Anomaly patterns detected