Chapter 06 · AI Agents AI Agents Track No-code

Paste a messy bug report, get a structured triage verdict in seconds.

Two visual LangFlow agents: a QA Buddy that answers testing questions over Groq, and a Bug Report Classifier whose engineered prompt assigns severity, category, priority, and reproduction confidence like a 15-year QA lead.

01 What students build

LangFlow is a drag-and-drop canvas for LLM pipelines. Flow 1 (QA Buddy) wires Chat Input straight into a Groq model with a QA-only persona. Flow 2 (Bug Classifier) inserts a Prompt Template between input and model: it defines four severity levels (CRITICAL, HIGH, MEDIUM, LOW), seven categories (UI/UX, FUNCTIONAL, PERFORMANCE, SECURITY, DATA, INTEGRATION, INFRASTRUCTURE), and forces a structured response with reproduction confidence and similar known issues. Both export as shareable JSON flows.

Visual Flow Canvas

Every component is a node on a canvas: chat input, prompt template, Groq model, chat output. Rewire the pipeline by dragging edges, no code needed.

Prompt Template Node

The classifier's brain is a single {bug_report} template encoding severity rules, category definitions, and the exact output sections to fill.

Structured Triage

Output is not free text: severity, category, priority, steps-clear check, environment check, and frequency come back in fixed sections you can route onward.

02 Architecture

Chat Inputraw bug report
──▶
Prompt Template{bug_report} + rules
──▶
Groq LLMQA lead persona
──▶
Chat Outputstructured verdict
flow 1 (qa buddy) skips the template · flow 2 (classifier) adds it · both ship as importable json

03 Live demo (simulated with sample data)

This is a UI walkthrough with pre-baked sample data so you can feel the classifier without running LangFlow. In the course you import the two JSON flows into LangFlow and connect your own Groq API key.

langflow · localhost:7860 demo · sample data

04 Inputs & outputs

What goes in

  • Raw bug reports pasted into the chat panel (any format)
  • QA and test-automation questions for the QA Buddy flow
  • Groq API key wired into the model node
  • Tunable prompt template: severity rules, categories, output sections

What comes out

  • Severity verdict: CRITICAL / HIGH / MEDIUM / LOW with rationale
  • Category out of 7 (FUNCTIONAL, PERFORMANCE, SECURITY, …)
  • Reproduction confidence: steps clear, environment, frequency
  • Similar known issues and suggested next owner
  • Exportable flow JSON your whole team can import

05 How it's built

Install and launch LangFlow, then open the visual canvas at localhost:7860.
Build the QA Buddy: Chat Input, Groq model with a Senior QA persona, Chat Output. Test it in the built-in playground.
Write the classifier prompt: severity ladder (CRITICAL to LOW), 7 category definitions, and a fixed response skeleton.
Drop a Prompt Template node with a {bug_report} variable and wire Chat Input into it.
Connect the filled prompt to the Groq model and route the verdict to Chat Output.
Export both flows as JSON so anyone can import, swap the API key, and run.

06 Tech stack

LangFlowGroq LLMPrompt Template nodeChat Input / OutputSeverity rubric promptJSON flow exportPlayground testing