☁️ Cloud Engine

The Cloud Engine is Vespera's infrastructure operations brain. It combines AI-generated recommendations, a RAG knowledge base of cloud best practices, compliance guardrails, and direct Terraform CLI validation into a single Discord workflow.

---

Overview

graph LR
    A[User: /cloud-advise] --> B[CloudAIAdvisor]
    B --> C[RAG Knowledge Base\n186 best practices]
    B --> D[Guardrails\nBudget · Security · Compliance]
    B --> E{Model Router}
    E -->|Fast| F[Groq Llama 3.3]
    E -->|Complex| G[Gemini Pro 1.5]
    F & G --> H[JSON Recommendation]
    H --> I[Discord Embed]

---

AI Model Routing

Task	Model	Reason
General recommendations	Groq Llama 3.3	Ultra-fast (~1–2s), sufficient reasoning
Complex doc analysis	Gemini Pro 1.5	1M+ token context window for log/state files
Balanced quality	Groq Mixtral	Trade-off between speed and depth

---

Key Features

RAG Knowledge Base

• 186 best practices ingested from Markdown across GCP, AWS, and Azure
• Hybrid search: keyword matching + metadata category filtering
• Categories: Security (48), Performance (47), Cost (46), Reliability (45)

Guardrails System

Runs before any AI generation to block or warn on bad requests:

Guardrail	Action
GPU workload + Low budget	⛔ BLOCKED
Healthcare data + Standard security	⚠️ WARNING
HIPAA compliance missing audit logging	⛔ BLOCKED
Overly complex architecture for beginners	⚠️ WARNING

Chain-of-Thought Reasoning

For every recommendation, the model is forced through a structured internal [Reasoning] block before generating output:

1. Requirement Analysis
2. Constraints Identification
3. Best Practice Matching
4. Trade-off Analysis (Cost / Simplicity / Security / Scalability)

This prevents the "commit-then-justify" bias where an LLM picks an answer first and rationalizes it second.

Terraform Validation Pipeline

graph LR
    A[Session Created] --> B[terraform fmt]
    B --> C[terraform init]
    C --> D[terraform validate]
    D --> E[terraform plan]
    E --> F[terraform show -json]
    F --> G[Discord Report]

---

Discord Commands

/cloud-advise

Get AI-powered cloud infrastructure recommendations.

Parameters:
  use_case        (required) What are you building?
  provider        GCP | AWS | Azure | Any (compare all)
  budget          Low (<$100/mo) | Medium | High (>$1000/mo)
  security_level  Standard | Enhanced | Compliance (HIPAA/GDPR/PCI)
  ai_model        Groq Llama 3.3 | Groq Mixtral | Gemini Pro | Gemini Flash

/cloud-validate

Runs the full Terraform validation pipeline against a session's generated configuration.

/cloud-deploy

Generates Terraform code from an advisory session and creates a deployment session ID.

---

Technical Highlight — Lazy Glossary Injection

~90% Token Reduction

Rather than injecting the entire Cloud or D&D glossary into every prompt, the system scans the user's input for known keywords first. Only terms that appear in the input are injected into the prompt.

• Input: "I need a Kubernetes cluster for my microservices"
• Injected: only the Kubernetes and microservices glossary entries
• Result: token consumption reduced ~90%, with higher accuracy (no irrelevant noise in context)

---

Performance Metrics

Model	Avg Response	Notes
Groq Llama 3.3	1–2s	Recommended default
Groq Mixtral	2–3s	Higher quality
Gemini Pro	3–5s	Large context tasks
Gemini Flash	1–3s	Fast Google alternative

Operation	Time
RAG hybrid search	< 50ms
Terraform fmt	< 1s
Terraform init (cached)	1–3s
Terraform plan	3–10s