Cost optimization strategies in Agentic Ai - Model Pipeline Trace

Model Pipeline - Cost optimization strategies

This pipeline shows how an AI agent learns to optimize costs by analyzing data, training a model to predict cost-saving actions, and improving its decisions over time.

Data Flow - 7 Stages

1Data Collection

10000 rows x 8 columns→Gather raw cost and usage data from various sources→10000 rows x 8 columns

Row example: { 'resource_id': 'R123', 'usage_hours': 120, 'cost': 300, 'region': 'us-east', 'service_type': 'compute', 'timestamp': '2024-05-01', 'optimization_flag': 0, 'previous_savings': 50 }

↓

2Data Preprocessing

10000 rows x 8 columns→Clean data, handle missing values, encode categorical variables→10000 rows x 10 columns

Added one-hot encoding for 'region' and 'service_type', filled missing 'cost' with median

↓

3Feature Engineering

10000 rows x 10 columns→Create new features like 'cost_per_hour' and 'usage_category'→10000 rows x 12 columns

'cost_per_hour' = cost / usage_hours, 'usage_category' = 'high' if usage_hours > 100 else 'low'

↓

4Train/Test Split

10000 rows x 12 columns→Split data into training (80%) and testing (20%) sets→Train: 8000 rows x 12 columns, Test: 2000 rows x 12 columns

Training set used to teach the model, test set to evaluate

↓

5Model Training

8000 rows x 12 columns→Train a decision tree model to predict cost-saving actions→Trained model object

Model learns patterns to suggest when to optimize resource usage

↓

6Model Evaluation

2000 rows x 12 columns→Evaluate model accuracy and loss on test data→Accuracy: 0.85, Loss: 0.35

Model correctly predicts cost-saving actions 85% of the time

↓

7Prediction

New data sample with 12 features→Model predicts if cost optimization action should be taken→Prediction: 0 or 1 (No/Yes)

Input: {usage_hours: 150, cost_per_hour: 2.5, ...} -> Output: 1 (optimize)

Training Trace - Epoch by Epoch

Loss
0.7 |****
0.6 |*** 
0.5 |**  
0.4 |*   
0.3 |    
     1 2 3 4 5 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	0.65	0.6	Model starts learning basic patterns
2	0.5	0.72	Loss decreases, accuracy improves
3	0.42	0.78	Model captures more complex relationships
4	0.38	0.82	Training converges, performance stabilizes
5	0.35	0.85	Final model achieves good accuracy

Prediction Trace - 4 Layers

Layer 1: Input Features

Layer 2: Decision Tree Splitting

Layer 3: Leaf Node Prediction

Layer 4: Thresholding

Model Quiz - 3 Questions

What happens to the data shape after one-hot encoding in preprocessing?

A Number of rows decreases

B Number of columns increases

C Number of columns decreases

D Number of rows increases

Key Insight

This visualization shows how cost optimization can be learned by an AI model through data preparation, training, and prediction. The model improves by reducing loss and increasing accuracy, enabling it to suggest cost-saving actions effectively.