A curated collection of autonomous AI agents, tool-calling systems, security guardrails, and rapid prototyping workflows built during the 5-Day AI Agents: Intensive Vibe Coding Course with Google hosted on Kaggle.
- Credential: 🔗 Kaggle Badge #108 — 5-Day AI Agents (Danny Riggleman)
- Official Course: 🎓 5-Day AI Agents: Intensive Vibe Coding Course with Google
- Organized By: Google & Kaggle
- Audience / Take the Course: Anyone interested in modern agentic AI architectures, tool calling with Gemini & Google ADK, and high-velocity "vibe coding" workflows can access the curriculum and hands-on tracks directly at the Kaggle Learn Guide.
This repository is organized into five progressive modules directly aligned with the official Google & Kaggle curriculum:
google-ai-agents-vibe-coding/
├── 01-agent-foundations/ # Day 1: ReAct reasoning loops & tool registry architecture
├── 02-tool-use-weather-assistant/ # Day 2: Google Agents CLI / ADK weather agent with function calling
├── 03-autonomous-expense-agent/ # Day 3: Background autonomous expense tracking & state management
├── 04-agent-security-and-evaluation/ # Day 4: Prompt injection defense, guardrails & ADK quality evaluation
├── 05-mcp-vibe-coding-orchestration/ # Day 5: Model Context Protocol (MCP) tool integration & production workflows
├── LICENSE # MIT License
├── pyproject.toml # Project metadata and dependencies
└── README.md
- Focus: Transitioning from static LLM prompt-response pairs to autonomous Reasoning + Action (ReAct) execution loops.
- Key Patterns: Structured session memory, step-by-step reasoning (
Thought -> Action -> Observation), and deterministic tool execution registries.
- Project:
weather-assistant - Focus: Building modular, tool-equipped agents using the Google Agent Development Kit (ADK) and
google-agents-cli. - Key Patterns: JSON schema generation, external API querying, parameter validation, and conversational synthesis.
- Project:
ambient-expense-agent - Focus: Background task automation, state persistence, and modular agent skills.
- Key Patterns: Autonomous trigger loops, expense categorization schemas, state management, and reproducible test suites.
- Focus: Hardening agents against adversarial threat vectors and executing automated quality benchmarks.
- Key Patterns: Direct/indirect prompt injection defense, output secret sanitization, Google ADK custom metrics (
eval_config.yaml), and 1–5 quality scoring.
- Focus: Connecting agents to standardized Model Context Protocol (MCP) tool servers for dynamic discovery and production scaling.
- Key Patterns: Dynamic tool discovery, execution journaling, confidence scoring, and high-velocity prototyping ("vibe coding").
- Agent Frameworks: Google Agent Development Kit (ADK), Google Agents CLI, ReAct Architecture, Model Context Protocol (MCP)
- Models: Gemini 2.0 Flash / Pro, Claude 3.5 Sonnet, Local LLMs (Ollama)
- Languages & Runtimes: Python 3.11+, UV, FastAPI, Pydantic, TypeScript
- Tooling & Testing: Pytest, ADK Eval Engine, WebSockets, RESTful APIs, Git
git clone https://github.com/xylaes/google-ai-agents-vibe-coding.git
cd google-ai-agents-vibe-coding# Using uv (recommended)
uv venv
source .venv/bin/activate # Or .venv\Scripts\activate on Windows
# Install dependencies
uv pip install -e .# Day 1: ReAct Loop Demo
python 01-agent-foundations/react_agent_loop.py
# Day 4: Security & Evaluation Benchmark
python 04-agent-security-and-evaluation/agent_evaluator_and_guardrails.py
# Day 5: MCP Orchestration Demo
python 05-mcp-vibe-coding-orchestration/mcp_client_orchestrator.pyDanny Riggleman
- UpWork Profile: upwork.com/freelancers/xylaes
- GitHub: @xylaes
- Kaggle Profile: kaggle.com/dannyriggleman
- Education: B.S. in Computer Science & Philosophy (West Virginia University)
This project is licensed under the MIT License.