FrootAI — From Root to Fruit
The uniFAIng glue that binds infrastructure, platform, and application The telescope and the microscope for AI architecture. See the big picture. Master the tiny details. Design with confidence.
Modules: 18 | Duration: 16–22 hours | Level: Beginner → Expert Audience: Cloud Architects, AI Engineers, Platform Engineers, DevOps, CSAs Scope: Everything AI — from a single token to a production agent fleet Last Updated: March 2026
What is FrootAI?
FrootAI = AI Foundations · Reasoning · Orchestration · Operations · Transformation
You are a cloud architect. You build platforms that host workloads. But the workloads have changed. Every application is becoming AI-native — language models, retrieval pipelines, autonomous agents, copilot integrations. You need to understand the entire tree, from the roots (how a token becomes a thought) to the fruit (a production agent that serves millions).
FrootAI gives you both lenses:
| 🔭 Telescope (Big Picture) | 🔬 Microscope (Tiny Details) |
|---|---|
| How does an AI Landing Zone fit into enterprise architecture? | What is the difference between top_k=40 and top_k=10? |
| When should I use Semantic Kernel vs Microsoft Agent Framework? | How does BPE tokenization split “unbelievable” into sub-tokens? |
| What hosting pattern works for multi-agent systems? | Why does temperature=0.0 still not guarantee determinism? |
| How do I design a RAG pipeline for 10M documents? | What is the cosine similarity threshold for relevant retrieval? |
“The soil is the platform. The roots are the fundamentals. The trunk is reasoning. The branches are orchestration. The canopy is operations. The fruit is transformation. You need to understand the entire tree to grow the right solutions.”
The FROOT Framework
FrootAI organizes everything in the GenAI world into five layers — each building on the last, each essential to the whole:
Module Map
🌱 F — Foundations (The Roots)
What AI is, how it thinks, the vocabulary you need
| # | Module | Duration | What You’ll Master |
|---|---|---|---|
| F1 | GenAI Foundations | 60–90 min | Transformers, attention, tokenization, inference, parameters (temperature, top-k, top-p), context windows, embeddings |
| F2 | LLM Landscape & Model Selection | 45–60 min | GPT, Claude, Llama, Gemini, Phi — benchmarks, open vs proprietary, when to use what |
| F3 | AI Glossary A–Z | Reference | Comprehensive glossary defined — from “ablation” to “zero-shot”. The dictionary you keep open in another tab |
🪵 R — Reasoning (The Trunk)
How to make AI think well — reliably, accurately, without hallucination
| # | Module | Duration | What You’ll Master |
|---|---|---|---|
| R1 | Prompt Engineering & Grounding | 60–90 min | System messages, few-shot, chain-of-thought, structured output, guardrails, function calling |
| R2 | RAG Architecture & Retrieval | 90–120 min | Chunking, embeddings, vector search, Azure AI Search, semantic ranking, reranking, hybrid search |
| R3 | Making AI Deterministic & Reliable | 60–90 min | Hallucination reduction, grounding techniques, temperature vs top-p tuning, evaluation metrics, guardrails |
🌿 O — Orchestration (The Branches)
Connecting AI components into intelligent systems — agents, tools, frameworks
| # | Module | Duration | What You’ll Master |
|---|---|---|---|
| O1 | Semantic Kernel & Orchestration | 60 min | Plugins, planners, memory, connectors, comparison with LangChain, when to use SK |
| O2 | AI Agents & Microsoft Agent Framework | 90–120 min | Agent concepts, planning, memory, tool use, AutoGen, multi-agent, deterministic agents |
| O3 | MCP, Tools & Function Calling | 60–90 min | Model Context Protocol, tool schemas, function calling patterns, A2A, MCP servers, registry |
🏗️ O — Operations (The Canopy)
Running AI in production — platforms, infrastructure, hosting, low-code
| # | Module | Duration | What You’ll Master |
|---|---|---|---|
| O4 | Azure AI Platform & Landing Zones | 60–90 min | AI Foundry, Model Catalog, deployments, endpoints, AI Landing Zone, enterprise patterns |
| O5 | AI Infrastructure & Hosting | 60–90 min | GPU compute, Container Apps, AKS, App Service, model serving, scaling, cost optimization |
| O6 | Copilot Ecosystem & Low-Code AI | 45–60 min | M365 Copilot, Copilot Studio, Power Platform AI, GitHub Copilot, extensibility |
🍎 T — Transformation (The Fruit)
Turning AI into real-world impact — safely, efficiently, at scale
| # | Module | Duration | What You’ll Master |
|---|---|---|---|
| T1 | Fine-Tuning & Model Customization | 60–90 min | When to fine-tune vs RAG, LoRA, QLoRA, RLHF, DPO, evaluation, MLOps lifecycle |
| T2 | Responsible AI & Safety | 45–60 min | Content safety, red teaming, guardrails, Azure AI Content Safety, evaluation frameworks |
| T3 | Production Architecture Patterns | 60–90 min | Multi-agent hosting, API gateway for AI, latency optimization, cost control, monitoring |
📋 Reference & Assessment
| # | Module | Duration | Purpose |
|---|---|---|---|
| REF | Quick Reference Cards | Reference | One-page cheat sheets for every concept — pin them to your wall |
| QUIZ | Quiz & Assessment | 20 min | 25 questions covering the full FrootAI curriculum |
How the FROOT Layers Connect
Every layer builds on the one below it. You can jump to any module — but understanding flows upward from roots to fruit:
Who is This For?
Learning Paths
Not sure where to start? Pick your path:
🚀 Path 1: “I’m New to AI” (6–8 hours)
Start from the roots, build understanding layer by layer
F1 → F3 → F2 → R1 → R2 → R3 → QUIZ⚡ Path 2: “I Need to Build an Agent NOW” (4–5 hours)
Fast-track to agent development with just enough foundation
F1 (Sections 1.1–1.5) → R1 → O2 → O3 → O1 → T3🏗️ Path 3: “I’m Designing AI Infrastructure” (5–6 hours)
Platform and operations focus for infra architects
F1 → O4 → O5 → T3 → R2 → T1🔍 Path 4: “I Need to Make AI Reliable” (3–4 hours)
Determinism, grounding, and safety — for when AI must not fail
R3 → R1 → R2 → T2 → REF🎯 Path 5: “The Complete Journey” (16–22 hours)
Every module, roots to fruit — become the AI architect
F1 → F2 → F3 → R1 → R2 → R3 → O1 → O2 → O3 → O4 → O5 → O6 → T1 → T2 → T3 → REF → QUIZThe FrootAI Promise
By the time you’ve worked through FrootAI, you will be able to:
✅ Explain how a transformer turns tokens into intelligence — in 60 seconds, to any audience
✅ Design RAG pipelines that retrieve the right information with the right relevancy scores
✅ Choose between Semantic Kernel, Microsoft Agent Framework, LangChain, and direct API calls
✅ Build deterministic AI systems that minimize hallucination and maximize grounding
✅ Architect AI Landing Zones with proper networking, security, and cost controls
✅ Deploy agents on Container Apps, AKS, or serverless — and know the tradeoffs
✅ Connect AI to external systems via MCP, function calling, and tool orchestration
✅ Evaluate when to fine-tune vs when to RAG — and how to do both
✅ Govern AI responsibly with content safety, red teaming, and guardrails
✅ Scale production AI systems with proper monitoring, cost control, and architecture patterns
Quick Navigation
🌱 F | 🪵 R | 🌿 O | 🏗️ O | 🍎 T |
🔌 FrootAI MCP Server
FrootAI is not just documentation — it’s a programmable knowledge base. Connect it to any MCP-compatible AI agent:
{
"mcpServers": {
"frootai": {
"command": "npx",
"args": ["frootai-mcp"]
}
}
}23 tools: 6 static (knowledge) + 4 live (Azure docs, MCP registries) + 3 agent chain (build → review → tune) + 3 AI ecosystem (model catalog, pricing, compare) + 6 compute
See mcp-server/README.md for full setup.
🖥️ FrootAI VS Code Extension
Standalone engine — works from any workspace, no clone needed:
Ctrl+Shift+X → Search "FrootAI" → Install17 commands: Init DevKit (.github Agentic OS), Init TuneKit, Init SpecKit, Auto-Chain Agents, MCP Install/Start/Configure, Search Knowledge, Look Up AI Term, and more.
4 sidebar panels: Solution Plays (20) · MCP Tools (22) · Knowledge Hub (knowledge modules) · AI Glossary (comprehensive glossary)
FrootAI v2.2 — The uniFAIng glue for AI architecture. From the roots to the fruits. Knowledge modules · MCP tools · solution plays · comprehensive glossary Built with 🌳 by the FrootAI community for the Azure community.