01 What students build
One catalog to rule them all: catalog/projects.json defines each project's system prompt, task prompt, inputs, outputs, workflow and sample input. A generator script (scripts/generate-flows.mjs) rebuilds the n8n and LangFlow JSON for all 13 projects from that single source of truth. The React studio (Vite, port 5177) runs the same catalog through a shared Node LLM wrapper (port 8787) that speaks Groq, OpenAI/ChatGPT, Anthropic, OpenRouter, Ollama and LM Studio. Agents range from a Test Strategy Builder and Selenium-to-Playwright Converter to a Flaky Test Detector, Self-Healing Automation Agent and Release Readiness scorer.
n8n Fleet
13 importable n8n workflow JSON files, one per project, regenerated from the catalog so prompts never drift between formats.
LangFlow Fleet
The same 13 projects as importable LangFlow flows for visual, drag-and-drop experimentation with every agent.
React Studio
A studio app with a project picker, provider switcher and sample inputs · run any agent against any of 6 LLM providers, local or cloud.
02 Architecture
03 Live demo (simulated with sample data)
This is a UI walkthrough with pre-baked sample data so you can feel the studio without installing anything. In the course you run the real thing: npm run dev:all starts the React app on 5177 and the LLM wrapper on 8787.
04 Inputs & outputs
What goes in
- A project pick from the 13-entry catalog (agents + tools)
- The project's sample input, or your own (code, logs, DOM, schema)
- A provider choice: Groq, OpenAI/ChatGPT, Anthropic, OpenRouter, Ollama, LM Studio
- API keys via
.env(none needed for local Ollama / LM Studio)
What comes out
- The agent's structured result: converted code, flake ranking, audit report and more
- Importable n8n workflow JSON per project
- Importable LangFlow flow JSON per project
- A portfolio piece you can demo in interviews, three ways
05 How it's built
catalog/projects.json: system prompt, task prompt, inputs, outputs, integrations, workflow and a realistic sample input each.generate-flows.mjs to rebuild every n8n and LangFlow JSON from the catalog, so one edit updates all formats./api/run surface that normalizes Groq, OpenAI, Anthropic, OpenRouter, Ollama and LM Studio behind a single call shape.npm run dev:all: studio on 5177, wrapper on 8787, keys from .env.