Intent-Driven QA Framework

An AI-assisted QA automation framework that separates test intent, test memory, and test execution to reduce repeated test design work and improve regression coverage.

This project demonstrates senior-level QA system design by treating testing as a decision system, not just a collection of scripts.

---

Core Idea

Manual tests define intent. Memory preserves context. Automation executes deliberately.

Traditional automation frameworks duplicate test thinking inside code. Over time, intent gets lost, regression coverage degrades, and engineers re-solve the same problems.

This framework explicitly separates responsibilities so that:

• Test intent remains human-owned
• Historical knowledge compounds over time
• Automation is written with clear purpose

---

Architecture Overview

manual-tests/        → Intent layer (source of truth)
qa-agent/            → Decision layer (AI-assisted planning)
automation/          → Execution layer (Playwright)
docs/                → System rules and design philosophy

Each layer has a single responsibility.

---

Intent Layer (Manual Tests)

Manual regression tests are written as structured markdown files.

They capture:

• Test intent (what behavior is validated)
• Feature and page context
• Risk areas
• Automation status

Manual tests are never duplicated inside automation code.

Example:

manual-tests/login/login-invalid-password.md

---

Decision Layer (QA Agent + Memory)

A lightweight QA Agent assists with test planning, not execution.

Memory Store

• Uses Chroma as a vector database
• Stores short semantic summaries (not full steps)
• Enriched with structured metadata

What the Agent Does

When a new Jira ticket arrives:

1. Ticket is summarized
2. Memory is queried for similar intent
3. Existing coverage is surfaced
4. Gaps and regression risks are identified

The agent provides recommendations, not commands.

---

Execution Layer (Playwright)

Playwright is used strictly as the execution engine.

Design Rules

• Page Objects contain selectors and actions only
• Assertions live only in tests
• Tests map one-to-one with manual intent

Example traceability:

// Manual Test: manual-tests/login/login-invalid-password.md

This preserves long-term explainability.

---

AI Tooling Philosophy

AI tools (Cursor, Copilot, agents) are used as execution assistants.

They are constrained so that AI:

• Does not invent test intent
• Does not add undocumented assertions
• Does not modify manual tests

Humans own intent. AI assists implementation.

---

Example Application

Automation targets AutomationExercise (public demo e-commerce site).

Chosen because it provides:

• Realistic authentication and checkout flows
• Stable selectors (data-qa)
• Sufficient complexity for portfolio demonstration

---

Why This Matters

This framework demonstrates:

• Intent preservation over time
• Regression-aware test planning
• Scalable automation architecture
• Responsible AI integration in QA

It reflects how senior SDETs and QA architects think about test systems.

---

Repository Status

• ✅ Manual intent defined
• ✅ Automation aligned to intent
• ✅ AI-assisted planning design documented
• 🚧 QA Agent implementation (planned)

---

Who This Is For

• Senior QA / SDET interviews
• Teams struggling with brittle automation
• Engineers exploring responsible AI usage in testing

---

Next Steps

• Implement QA Agent ingestion pipeline
• Add additional feature coverage
• Integrate CI execution

---

Testing is not about scripts. It is about preserving intent and judgment over time.