<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> This repository offers a hands-on workshop for building production-grade AI agent systems from scratch, focusing on practical implementation of multi-agent architectures and advocating for simpler, effective designs. Aimed at AI engineers and technical leads, it provides a deep dive into agentic workflows, tool use, and system design, enabling rapid skill development and project deployment.

How It Works

The workshop employs the Model Context Protocol (MCP) framework, specifically FastMCP, to orchestrate two core AI agents: a Deep Research Agent leveraging Gemini with Google Search and YouTube analysis, and a LinkedIn Writing Workflow featuring an evaluator-optimizer loop. The design philosophy prioritizes the "simplest system that reliably solves the problem," contrasting complex multi-agent setups with more efficient single-agent-plus-tools approaches. Key patterns include tool-use agents, grounded search, structured LLM output via Pydantic, and LLM-as-judge evaluation.

Quick Start & Requirements

• Installation: Clone the repository, copy .env.example to .env and add your GOOGLE_API_KEY (and optional OPIK_API_KEY), then run uv sync to install dependencies. A make test-end-to-end command verifies the setup.
• Prerequisites: Python 3.12+, uv 0.7+, GNU Make, and a Google Cloud AI Studio API key for Gemini. An Opik account is recommended for advanced observability and evaluation.
• Resources: The workshop is designed for 2-4 hours of engagement, with running the finished code taking approximately 30 minutes.
• Links: Full workshop video on YouTube, slides, and a related Agentic AI Engineering Course with a Discord community are available.

Highlighted Details

• Deep Research Agent: Integrates Gemini with Google Search grounding and native YouTube video analysis for comprehensive topic exploration.
• LinkedIn Writing Workflow: Utilizes an iterative evaluator-optimizer loop for generating and refining LinkedIn posts, culminating in image generation.
• LLM-as-Judge Evaluation: Implements an automated quality scoring pipeline using Opik to measure agent performance against defined criteria.
• Harness Engineering: Teaches principles of building robust AI environments beyond prompting, including tools, state management, guardrails, and testing.

Maintenance & Community

The project is maintained by Louis-François Bouchard, Paul Iusztin, and Samridhi Vaid. A Discord community is associated with the broader Agentic AI Engineering Course, offering access to experts and fellow learners.

Licensing & Compatibility

The project is released under the MIT License, permitting broad use, modification, and distribution, including for commercial purposes.

Limitations & Caveats

The project strongly advocates for simpler AI architectures, cautioning against premature or unnecessary multi-agent complexity. The "implement yourself" mode requires careful setup within a scoped directory to prevent agents from accessing reference solutions, ensuring a genuine build experience. A Google API key is required for core LLM functionality.