I gave a talk in the QA track at the community event “福岡Tech LT大忘年会” held in Fukuoka.
In this article, I’ll share the content of my presentation along with the slides.
Presentation Content
Below are excerpts from the presentation slides Giselleで作るAI QAアシスタント 〜 Pull Requestレビューに継続的QAを.
Presentation Overview
I introduced our initiative of building an AI QA assistant using Giselle to implement continuous QA during Pull Request (PR) reviews.
- A mechanism to integrate QA into PR reviews
- Customizing QA perspectives for specific projects
- Effects and learnings from implementation
Background: QA Challenges
- QA right before release leads to high rework costs
- Coming up with QA perspectives is tedious → Tendency to only cover regular cases
- Review load increases as PRs grow
「QA観点出しのハードルを下げたい…」
(“I want to lower the barrier to coming up with QA perspectives…”)
What is an AI QA Assistant?
PRの変更内容を自動で分析し、QA観点でフィードバックを提供するAIエージェント
An AI agent that automatically analyzes PR changes and provides feedback from a QA perspective
- Identifies the impact scope of changes
- Suggests testing perspectives
- Points out edge cases
- Detects regression risks
Why We Chose Giselle
- Build AI workflows with no-code
- Native GitHub integration - naturally fits into the development flow
- Freely customizable QA perspectives
- Open source (OSS) with transparency
- Because we’re developing it ourselves!
giselles.ai / GitHub - giselles-ai/giselle
Integrating QA into PR Reviews
Create PR on GitHub
↓
Giselle workflow auto-triggers
↓
Generate review from QA perspective
↓
Feedback as PR comment
- Integrate QA without changing existing development flow
- Engineers just create PRs as usual
- QA feedback is completed within the PR
Customizing QA Perspectives
Analyzes PR title, body, and diff to generate two types of output
1. Manual QA Checklist
- Generates checklist from Happy Path Testing perspective
- Specific test items anyone can understand
2. Prompt for AI Agents
- E2E test generation prompt for Playwright MCP
- Directly usable with AI coding tools like Claude Code, Codex, Cursor, etc.
Implementation Effects
Lowered the barrier to QA
- AI provides the perspectives → Just QA based on them
- Reduced burden of thinking “what should be tested”
- Also presents edge cases and regression risks
「QA観点出しが面倒でレギュラーケースだけだったのが、AIの提案をベースにしっかりQAできるようになった！」
(“QA perspective generation used to be tedious so we only covered regular cases, but now we can properly QA based on AI suggestions!”)
Learnings and Future
- AI isn’t perfect, but effective as a safety net for preventing oversights
- Continuous improvement of prompts is important
- Became a catalyst for fostering QA culture across the team
Presentation Summary
- Built an AI QA assistant using Giselle
- Integrate QA into PR reviews
- Customize QA perspectives for specific projects
- Reduce rework costs through early feedback
Closing
Thank you to everyone who attended, and to the organizers: KINTOテクノロジーズ株式会社, 株式会社トライアルカンパニー, 株式会社ヌーラボ, 株式会社マネーフォワード, and LINEヤフー株式会社!
The Fukuoka Tech LT Year-End Party featured many stimulating presentations from other speakers, and I could feel the energy of Fukuoka’s tech community firsthand.
That’s all from the Gemba, where we’re making full use of the AI QA assistant built with Giselle.
References