ES
joseph
Say hi ↗
MEASURE • PREDICT • SAVE • MEASURE • PREDICT • SAVE •
case 05 / 05

Aura Optimizer 

Machine learning model that predicts AWS cloud spend and suggests scaling optimizations.

Year
2025
Role
Data · Cloud FinOps
Status
Proprietary source code
( architecture )

How it's built

CloudWatchCost ExplorerPythonScikit-LearnS3AWS SDK
train≡EventBridgenightly↗CloudWatchmetrics↗Cost ExplorerbillingλCollectorλ python◉S3 Data Lakeparquet✦Forecast Modelscikit-learnλRecommenderλ lambda◉DynamoDBactions▭Slackalerts
Hover a service to trace its connections
( the problem )

Cloud bills only surprised the team at the end of the month, when it was too late to right-size anything.

( the build )

A nightly pipeline collects usage and cost signals, forecasts the month with a gradient-boosted model and posts concrete right-sizing actions to Slack.

−31%
monthly AWS spend
±4%
forecast error
nightly
fully automated
( request flow )
  1. 01EventBridge fires a nightly collection job.
  2. 02A collector Lambda pulls CloudWatch metrics and Cost Explorer data into an S3 lake.
  3. 03A Scikit-Learn model forecasts spend and finds idle or oversized resources.
  4. 04Recommendations are stored in DynamoDB and posted to Slack with one-click actions.
aura.console● live
−31%
monthly AWS spend
±4%
forecast error
nightly
fully automated
handler.py
1model = GradientBoostingRegressor()
2model.fit(X_usage, y_cost)
3forecast = model.predict(next_30d)
4recs = rightsize(idle(instances))
5slack.post(recs)
6▍
main ✓Python · UTF-8 · Ln 6
Next case · 01
Nexus AI Content Engine →