Kaggle AI Agents Intensive Course Expert

This skill provides expert guidance on the Kaggle 5-Day AI Agents Intensive course materials and the VaaS (Vendor-as-a-Service) capstone project.

Course Structure

The repository contains complete materials for all 5 days:

Day 1: Introduction to Agents

Location: day1_intro_to_agents/

• day-1a-from-prompt-to-action.ipynb - Basic agent creation
• day-1b-agent-architectures-28a9ac.ipynb - Multi-agent systems

Key Concepts:

• Agent vs LLM (agents take actions, not just respond)
• Sequential workflow patterns
• Multi-agent orchestration
• output_key pattern for state passing
• LlmAgent with tools and sub-agents

Day 2: Agent Tools

Location: day2_agent_architecture/

• day-2a-agent-tools.ipynb - Custom function tools
• day-2b-agent-tools-best-practices.ipynb - MCP integration, long-running operations

Key Concepts:

• Custom function tools with type hints
• FunctionTool wrapper
• AgentTool for agent-as-tool pattern
• Model Context Protocol (MCP) integration
• McpToolset for external services
• Long-running operations with ToolContext
• request_confirmation() for human-in-the-loop
• App with ResumabilityConfig

Day 3: Sessions & Memory

Location: day3_tools_and_memory/

• day-3a-agent-sessions.ipynb - Session management
• day-3b-agent-memory-with-answers.ipynb - Long-term memory

Key Concepts:

• InMemorySessionService vs DatabaseSessionService
• Session state with tool_context.state
• EventsCompactionConfig for context management
• Persistent workflows across restarts
• Memory systems for long-term knowledge

Day 4: Evaluation & Observability

Location: day4_evaluation_and_scaling/

• day-4a-agent-observability.ipynb - Logging, traces, metrics
• day-4b-agent-evaluation.ipynb - Testing and evaluation

Key Concepts:

• LoggingPlugin for standard traces
• Custom plugins with callbacks
• before_tool_callback and after_tool_callback
• adk web --log_level DEBUG for debugging
• ADK web UI for trace inspection
• Production observability patterns

Day 5: A2A Protocol & Deployment

Location: day5_capstone/

• day-5a-agent2agent-communication.ipynb - Critical for VaaS understanding
• day-5b-agent-deployment.ipynb - Agent Engine deployment

Key Concepts:

• Agent2Agent (A2A) protocol for cross-organizational communication
• RemoteA2aAgent (consumer side)
• to_a2a() (provider side)
• Agent Card at /.well-known/agent-card.json
• Streaming endpoints
• Framework-agnostic interoperability
• Vertex AI Agent Engine deployment

VaaS Capstone Project Concept

Location: competition_description_final.md

Core Innovation

VaaS (Vendor-as-a-Service) solves the Innovation-Adoption Gap:

• Small developers build great AI tools quickly ("vibe coding")
• Enterprises cannot consume them (no SOC2, GDPR, liability insurance)
• Traditional SaaS requires vendor certification ($100K+)

Revolutionary Insight

"A small developer's app does not need to be 'enterprise-ready' if the enterprise controls what leaves their environment."

How VaaS Works

Enterprise (ADK agents)
    ↓
    Security Filter (masks PII)
    ↓
    RemoteA2aAgent (A2A boundary)
    ↓
    to_a2a() (CrewAI vendor)
    ↓
    Vendor sees only sanitized data

Business Model Shift:

• Vendor = capability provider (NOT data processor)
• Enterprise = data controller (filtering stays internal)
• A2A = boundary enforcement

Five Barriers VaaS Removes

1. Data leakage risk
2. Enterprise compliance hurdles
3. Vendor liability concerns
4. Integration costs (standardized A2A)
5. Time-to-adoption barriers

Key Technical Achievement

Cross-framework A2A integration:

• Enterprise: Google ADK agents
• Vendor: CrewAI agent
• Communication: A2A protocol

This proves VaaS is framework-agnostic and universally applicable.

When to Use This Skill

Use this skill when users ask about:

• ✅ "How do I use agents?" → Day 1 materials
• ✅ "What are MCP tools?" → Day 2b
• ✅ "How do sessions work?" → Day 3a
• ✅ "How do I debug agents?" → Day 4a
• ✅ "What is A2A protocol?" → Day 5a
• ✅ "What is VaaS?" → Competition description
• ✅ "How does the capstone work?" → VaaS architecture
• ✅ "Cross-framework agents?" → ADK ↔ CrewAI via A2A
• ✅ "Enterprise AI adoption?" → Innovation-Adoption Gap

Key Commands

Navigate to Course Materials

cd day1_intro_to_agents  # Day 1
cd day2_agent_architecture  # Day 2
cd day3_tools_and_memory  # Day 3
cd day4_evaluation_and_scaling  # Day 4
cd day5_capstone  # Day 5

Read Key Files

# Competition submission
cat competition_description_final.md

# Course notebooks (use Read tool)
Read: day5_capstone/day-5a-agent2agent-communication.ipynb

Search for Concepts

# Find A2A examples
grep -r "RemoteA2aAgent" day5_capstone/

# Find tool examples
grep -r "FunctionTool" day2_agent_architecture/

Important Context

Repository Purpose

This repository contains:

1. Course materials: Complete Jupyter notebooks for Days 1-5
2. Competition submission: VaaS capstone description
3. Notes: Personal annotations and learning summaries

Related Repository

enterprise-gov-docs-a2a-capstone contains the actual implementation:

• Multi-agent system (IntakeAgent, ProcessingAgent)
• Security layer (PII filtering with 7 patterns)
• A2A integration (RemoteA2aAgent ↔ CrewAI vendor)
• Demo: Spanish birth certificate translation

Competition Context

Kaggle AI Agents Intensive Capstone

• Submission URL: https://www.kaggle.com/competitions/agents-intensive-capstone-project
• Word limit: 1500 words
• Judging: Style, Clarity, Originality
• Focus: Business innovation + technical mastery

Examples

Example 1: Understanding A2A Protocol

User asks: "How does A2A work?"

Response approach:

1. Read day5_capstone/day-5a-agent2agent-communication.ipynb
2. Explain consumer side: RemoteA2aAgent points to Agent Card
3. Explain provider side: to_a2a() exposes agent capabilities
4. Show VaaS use case from competition description

Example 2: Implementing Custom Tools

User asks: "How do I create a tool?"

Response approach:

1. Reference Day 2a notebook
2. Show FunctionTool wrapper pattern
3. Explain type hints and docstrings
4. Point to security_filter example in VaaS capstone

Example 3: Session Management

User asks: "How do I save agent state?"

Response approach:

1. Reference Day 3a notebook
2. Compare InMemorySessionService vs DatabaseSessionService
3. Show session state usage: tool_context.state["key"] = value
4. Explain when to use context compaction

Example 4: VaaS Business Model

User asks: "What problem does VaaS solve?"

Response approach:

1. Read competition_description_final.md
2. Explain Innovation-Adoption Gap
3. Show five barriers removed
4. Explain business model shift (capability provider vs data processor)
5. Highlight cross-framework achievement

Best Practices

1. Always check the actual notebooks - Use Read tool on .ipynb files for accurate code examples
2. Reference Day 5 for A2A - It's the breakthrough that enables VaaS
3. Explain VaaS holistically - It's not just tech, it's a business model
4. Show cross-framework value - ADK ↔ CrewAI proves framework-agnostic interoperability
5. Connect concepts - Show how Days 1-4 build foundation for Day 5

Troubleshooting

"I can't find the capstone implementation" → It's in a separate repo: enterprise-gov-docs-a2a-capstone

"What's the difference between this repo and the capstone repo?" → This repo = course materials + competition description → Capstone repo = actual working implementation

"How do I explain VaaS quickly?" → "Small developers can sell to enterprises by letting the enterprise control what data crosses the A2A boundary. The vendor provides capabilities, not data processing."

Version History

• v1.0 (2025-01-17): Initial skill creation with complete Day 1-5 coverage and VaaS capstone understanding