Chapter 04 · AI Agents AI Agents Track Full-stack Project

Point it at a Jira ticket, get a formal test plan a lead would sign off.

The first real agent of the track: it authenticates to Jira (or Azure DevOps), fetches the ticket, converts the rich-text description into a clean payload, and has an LLM acting as a Principal QA Architect write a complete test plan you can download and share.

01 What students build

A React frontend over a Node.js Express API, deployable to Vercel. The agent follows written SOPs (Standard Operating Procedures) for each tool: Jira fetch, LLM generation, and document rendering. Connections are testable before use, the provider switch covers local Ollama plus Groq and Grok in the cloud, and the output follows a real test plan template: objectives, scope, strategy, risks, entry and exit criteria.

Tracker Integration

Basic-auth clients for Jira REST and Azure DevOps work items, with Test Connection buttons so credentials fail loudly before generation starts.

SOP-driven Agent

Each step has a written SOP: how to fetch, how to parse Jira's ADF rich text, how to prompt, how to render. The agent is auditable, not magic.

Template Discipline

The LLM is constrained to the sections of a real test plan template, so every run produces the same reviewable structure, downloadable as a document.

02 Architecture

Ticket IDVWO-48 · creds
──▶
Jira / ADO fetchREST · basic auth
──▶
PayloadADF → plain text
──▶
LLMOllama · Groq · Grok
──▶
Test plantemplated Markdown
──▶
Download.docx / .md · React UI
jira fetch sop · llm generation sop · document generation sop

03 Live demo (simulated with sample data)

This is a UI walkthrough with pre-baked sample data so you can feel the agent without Jira credentials. In the course you run the real thing: React + Express locally or on Vercel, wired to your own Jira project and LLM.

test-planner · localhost:3000 demo · sample data

04 Inputs & outputs

What goes in

  • Jira base URL, email and API token (or an Azure DevOps org + PAT)
  • A ticket ID such as VWO-48, fetched live with title, description, acceptance criteria
  • Provider choice: Ollama local, Groq or Grok, with model string
  • Optional additional testing context for the prompt

What comes out

  • A complete test plan in Markdown following the template sections
  • Objectives, in/out scope, strategy, risks with mitigations, entry/exit criteria
  • A downloadable document (.docx via template, .md fallback)
  • Clear API errors for bad credentials or missing tickets

05 How it's built

Blueprint first: define the JSON schemas for the intermediate ticket payload and the plan request.
Link phase: prove the Jira and LLM connections with tiny scripts and Test Connection endpoints before building features.
Write the SOPs: jira_fetch_sop, llm_generation_sop and document_generation_sop as the agent's contract.
Implement the Express API: fetch the ticket via REST, convert the ADF description to plain text, handle 401/404 cleanly.
Prompt the LLM as a Principal QA Architect constrained to the test plan template, with generous timeouts for long generations.
Render the Markdown in the React UI, offer the document download, and deploy to Vercel.

06 Tech stack

React + ViteNode.js · ExpressJira REST APIAzure DevOps APIOllama · Groq · Grokmarked (Markdown).docx templatingVercel deploy