Infra Intelligence

Infra Intelligence — Infrastructure Department Intelligence Engine

Automated cloud cost monitoring, capacity forecasting, and optimization recommendations. UAOP Stage 1 + Stage 5 for Infrastructure.

When to use

• Monthly cloud cost review — "where's our money going?"
• After cost spikes — "why did our bill jump 30%?"
• Capacity planning — "when do we need to scale up?"
• Right-sizing review — "are we over-provisioned?"
• Before infrastructure changes — "what's the cost impact?"

When NOT to use

• Active incident response (use anomaly-detector + incident-responder)
• Security scanning (use threat-intelligence-feed + infrastructure-scanning)
• Terraform changes (use devops-patterns)

Pipeline

Step 1: Cost Data Collection

Collect spending data across AWS services:

COST CENTERS:
  Compute:
    - ECS Fargate tasks (backend + frontend)
    - Task CPU/memory allocation vs actual usage
    - Running hours per environment (dev/staging/production)

  Database:
    - RDS PostgreSQL instance class + storage
    - Read/write IOPS vs provisioned
    - Backup storage costs
    - Connection count vs max_connections

  Cache:
    - ElastiCache Redis node type
    - Memory utilization vs allocated
    - Eviction rate (over-provisioned if zero, under if high)

  Networking:
    - CloudFront data transfer
    - NAT Gateway costs (often the silent killer)
    - Cross-AZ data transfer

  Storage:
    - ECR image storage
    - S3 bucket storage + request costs
    - EBS volumes (if any)

  CI/CD:
    - GitHub Actions minutes consumed
    - ECR image push/pull bandwidth

  LLM:
    - API token usage by provider (OpenAI, Anthropic, Kimi)
    - Cost per request by model
    - Token usage trends

Step 2: Anomaly Detection

ANOMALY RULES:
  - Day-over-day increase >20% on any cost center → ALERT
  - Service cost exceeds monthly budget allocation → ALERT
  - Resource utilization <30% for >7 days → RIGHT-SIZE candidate
  - Resource utilization >80% for >3 days → SCALE-UP candidate
  - New cost center appears (unexpected service activated) → INVESTIGATE
  - LLM cost per request increasing without traffic increase → MODEL ISSUE

Step 3: Capacity Forecasting

FORECAST MODEL:
  For each resource:
    1. Collect 90 days of utilization data
    2. Fit linear + seasonal trend
    3. Project when resource hits 80% capacity (scale-up trigger)
    4. Estimate cost at projected usage levels

  Output:
    | Resource | Current Usage | 80% Threshold | Days Until 80% | Monthly Cost at 80% |
    |----------|--------------|---------------|----------------|-------------------|

Step 4: Optimization Recommendations

# Infrastructure Cost Report — [Month] [Year]

## Spending Summary
| Cost Center | This Month | Last Month | Delta | % of Total |
|------------|-----------|-----------|-------|-----------|
| Compute (ECS) | $[X] | $[X] | [+/-$X] | [X]% |
| Database (RDS) | $[X] | $[X] | [+/-$X] | [X]% |
| Cache (Redis) | $[X] | $[X] | [+/-$X] | [X]% |
| Networking | $[X] | $[X] | [+/-$X] | [X]% |
| LLM APIs | $[X] | $[X] | [+/-$X] | [X]% |
| **Total** | **$[X]** | **$[X]** | **[+/-$X]** | 100% |

## Anomalies Detected
[List of cost spikes with root cause analysis]

## Optimization Opportunities
| # | Recommendation | Savings/Month | Effort | Risk |
|---|---------------|--------------|--------|------|
| 1 | Right-size [resource] from [current] to [recommended] | $[X] | [S/M/L] | [Low/Med/High] |

## Capacity Forecast
[Table of resources approaching capacity limits]

## LLM Cost Breakdown
| Model | Requests | Tokens | Cost | Cost/Request |
|-------|---------|--------|------|-------------|

Step 5: Self-Improvement Loop

MEASURE:
  - Forecast accuracy (predicted cost vs actual)
  - Optimization recommendation ROI (savings achieved)
  - Anomaly detection precision (true vs false alerts)
  - Time from anomaly detection to resolution

REINFORCE:
  - Forecasting models that predicted accurately → use same method
  - Optimizations that saved money without incidents → template as standard
  - Cost centers with stable spending → reduce monitoring frequency

REGENERATE:
  - Update baselines after infrastructure changes
  - Add new cost centers when services are added
  - Adjust anomaly thresholds based on false positive rate

Cadence

DAILY:    Cost anomaly scan (>20% day-over-day)
WEEKLY:   Cost summary + right-sizing candidates
MONTHLY:  Full cost report + optimization recommendations + capacity forecast
QUARTERLY: Infrastructure strategy review + budget reforecast

Tools

Agents