Turntabl Agentic AI Masterclass

A two-day hands-on masterclass for software engineers on building multi-agent AI systems with CrewAI. The capstone is an autonomous crew of AI agents that triages GitHub issues, writes real code, opens pull requests, and reviews them — all automatically.

---

Repository Structure

turntablMasterClass/
├── curriculum/              # Masterclass teaching materials
│   ├── day1.md              # Day 1: Foundations of Agentic AI
│   ├── day2.md              # Day 2: Multi-Agent Systems & Production Deployment
│   ├── Agentic AI.png       # Diagram: how tool calling actually works
│   └── resources.md         # Additional reading and references
│
├── project/                 # Day 2 capstone — crew-based implementation
│   ├── agents/              # The four agents (PM, Team Lead, Frontend, Backend)
│   ├── tools/               # Custom GitHub and git tools
│   ├── .github/workflows/   # GitHub Actions workflow for automated triage
│   └── main.py              # Entry point: assembles and runs the crew
│
├── project-flows/           # Alternative implementation using CrewAI Flows
│   ├── crews/               # Specialized crews (triage, frontend, backend, review)
│   ├── flow.py              # Flow-based orchestration with deterministic routing
│   └── main.py              # Entry point
│
├── masterclass-ad-email.md  # Announcement email copy
└── masterclass-description.md

---

The Curriculum

The curriculum/ directory contains everything taught during the masterclass.

• Day 1 — Foundations of Agentic AI Covers the core concepts: what makes a system "agentic", the agent loop (Perceive → Reason → Act → Observe), how tool calling actually works under the hood, CrewAI's three pillars (Role, Goal, Backstory), prompt engineering for agents, and debugging common failure modes. Participants build their first single-agent GitHub issue analyzer.

• Day 2 — Multi-Agent Systems & Production Deployment Goes from one agent to a full team. Covers hierarchical delegation, task pipelines, shared context via task outputs, deploying to GitHub Actions, evaluating non-deterministic systems, and CrewAI Flows for deterministic orchestration. Participants build the multi-agent crew in project/.

---

The Project — Crew-Based Implementation

The project/ directory contains a multi-agent system with four agents that collaboratively triage and implement GitHub issues:

Agent	Role	LLM
Project Manager	Delegates tasks in the hierarchical crew	openai/gpt-4o
Technical Team Lead	Triages issues and reviews PRs	openai/gpt-4o
Senior Frontend Developer	Implements frontend changes, opens PRs	anthropic/claude-sonnet-4-20250514
Senior Backend Developer	Implements backend changes, opens PRs	anthropic/claude-sonnet-4-20250514

Workflow: Team Lead triages the issue → PM delegates implementation to the right dev → Dev clones the repo, writes code, and opens a PR → Team Lead reviews and approves → Team Lead posts a summary on the issue.

Running Locally

Prerequisites: Python 3.11+, uv installed.

# 1. Clone and enter the project
git clone https://github.com/blitzblade/turntablMasterClass.git
cd turntablMasterClass/project

# 2. Install dependencies
uv sync

# 3. Configure environment variables
cp .env.example .env
# Edit .env and set:
#   ANTHROPIC_API_KEY   — for the dev agents
#   OPENAI_API_KEY      — for the PM and Team Lead agents
#   GITHUB_TOKEN        — a PAT with repo access
#   GITHUB_REPO         — the target repo (e.g. "blitzblade/crew-commerce")

# 4. Run the crew on a specific issue
uv run python main.py --issue 1

Deploying as a GitHub Actions Workflow

The workflow at project/.github/workflows/agent-triage.yml automatically runs the crew whenever a new issue is opened (or labeled needs-triage).

1. Copy the project into the target repo — the workflow file needs to live in the repo you want the agents to work on.

2. Add repository secrets (Settings → Secrets and variables → Actions):

   • ANTHROPIC_API_KEY — your Anthropic API key
   • OPENAI_API_KEY — your OpenAI API key
   • GITHUB_TOKEN is provided automatically by Actions

3. Create an issue in the repo. The workflow triggers on issues: [opened, labeled], spins up the crew, and the agents will comment, label, and open a PR automatically.

Safety guardrails already configured in the workflow:

• timeout-minutes: 10 prevents runaway agents
• concurrency group cancels duplicate runs for the same issue
• max_rpm=5 in the crew config keeps API usage within rate limits

---

The Project — Flow-Based Alternative

The project-flows/ directory contains the same functionality implemented with CrewAI Flows instead of a single hierarchical crew. Flows add deterministic Python routing on top of crew executions, making the pipeline more inspectable and testable. See project-flows/README.md for details.

Compare the two approaches side by side to understand when to use each.

---

License

MIT — built for the Turntabl Agentic AI Masterclass.