DEV Community

Rick Wise
Rick Wise

Posted on

I Built an AI-Powered AWS Waste Detector (and Found $4,200 in My Own Account)

The $4,200 Mistake

I once left a staging environment running on AWS for 6 months.

Cost: $4,200

Traffic: 0

Users: 0

Value: 0

Just... existing. Racking up charges. Living its best life.

When I finally discovered it, I felt like I'd been paying rent for an apartment I forgot I had. 🤡

The Problem with AWS Cost Management

AWS Cost Explorer is powerful, but it's not designed to find waste. It shows you:

  • What you're spending
  • Where you're spending it
  • How spending trends over time

But it doesn't tell you:

  • Which resources are doing nothing
  • Which resources you forgot about
  • Which resources are over-provisioned
  • Where you're wasting money

What I Built

I just shipped AI-Powered Waste Detection for CloudWise.

It scans your AWS account and finds:

1. Orphaned Resources

  • RDS databases with zero connections
  • EBS volumes attached to nothing
  • Elastic IPs not attached to instances
  • Load balancers serving zero traffic

2. Idle Resources

  • EC2 instances with <5% CPU usage
  • RDS databases with zero queries
  • NAT Gateways with zero traffic
  • ELBs with zero requests

3. Over-Provisioned Resources

  • t3.2xlarge instances running cron jobs
  • RDS instances with 90%+ free storage
  • EC2 instances with excessive memory
  • Lambda functions with over-allocated memory

4. Cost Black Holes

  • Cross-AZ data transfer fees
  • S3 buckets in expensive storage classes
  • Old EBS snapshots
  • Unused Elastic IPs

How It Works

  1. Connect your AWS account (read-only IAM role, takes 2 minutes)
  2. AI scans your resources (analyzes usage patterns, costs, configurations)
  3. Get step-by-step recommendations (exactly what to do to fix each issue)

Example Output

🚨 Orphaned RDS Database Detected

Resource: prod-staging-db Cost: $247.32/month Last Connection: 187 days ago Status: Running but unused

Recommendation:

  1. Take final snapshot

  2. Delete database

  3. Estimated savings: $247.32/month ($2,967.84/year)

Risk Level: Low (no connections in 6 months)

Technical Implementation

Stack:

  • Frontend: React + TypeScript + TailwindCSS
  • Backend: Python + FastAPI
  • Database: PostgreSQL
  • AI: OpenAI GPT-4 for recommendation generation
  • AWS Integration: Boto3 + Cost Explorer API

Key Challenges:

  1. IAM permissions - Balancing security with functionality
  2. Cost attribution - Mapping resources to actual costs
  3. Usage pattern detection - Distinguishing idle from low-usage
  4. Recommendation quality - Avoiding false positives

What I Learned

1. Most AWS Waste is Obvious (Once You Look)

  • 70% of waste is orphaned or idle resources
  • 20% is over-provisioning
  • 10% is configuration issues (cross-AZ, storage classes)

2. The Problem Isn't Finding Waste, It's Looking

  • AWS Cost Explorer is intimidating
  • People avoid looking at their bills
  • "Out of sight, out of mind" mentality

3. Step-by-Step Instructions Matter

  • Knowing you have waste isn't enough
  • People need exact steps to fix it
  • Reducing friction = higher action rate

Results So Far

In my own AWS account:

  • Total waste found: $4,247/month
  • Orphaned resources: $1,847/month (RDS, EBS, ELB)
  • Idle resources: $1,200/month (EC2, RDS)
  • Over-provisioning: $800/month (EC2, Lambda)
  • Cross-AZ transfer: $400/month

Try It Yourself

I just launched this publicly today.

Try CloudWise: https://cloudcostwise.io

Free to use, no credit card required. Takes 2 minutes to connect your AWS account.

Would love your feedback, especially from DevOps engineers and platform teams!

What's Next

Coming soon:

  • Per-namespace cost tracking for Kubernetes/EKS
  • Automated remediation (with approval)
  • Slack/Teams alerts for cost anomalies
  • Multi-account support for organizations

Have you ever found unexpected AWS waste? Share your horror stories in the comments! 👇

P.S. That $4,200 staging environment? I could've bought a really nice espresso machine instead. Still thinking about it. ☕

Top comments (0)