Auravis is a learning and AI engineering portfolio project exploring a real software-delivery problem: business requirements often pass through several manual QA hand-offs before teams know whether a release actually satisfies the intended behavior. The target state is an autonomous, explainable UAT workflow with strong Java controls and evidence at every step.
Teams manually interpret BRDs, PRDs and user stories, then convert them into test scenarios. Auravis aims to automate that translation while preserving traceability to the original requirement.
Regression and UAT execution consume significant engineering time. Auravis uses deterministic Playwright execution so generated plans can become repeatable browser tests.
Test cases, screenshots, failures and release decisions often live in different places. Auravis persists mission history, execution results and evidence in one product flow.
A failed automation step still needs diagnosis. Auravis separates technical automation failures from potential product defects and builds toward controlled self-healing.
QA knowledge can remain in documents or individual experience. RAG and persisted project knowledge are used to ground future requirement and test reasoning.
The long-term goal is to move quality feedback earlier and ultimately let CI/CD consume an evidence-backed release recommendation.
Traditional flow
Autonomous UAT flow
What Auravis is intended to become
A user provides a complete business requirement and target environment; Auravis owns the downstream UAT workflow with minimal manual intervention.
Requirement, knowledge, test-planning, execution, diagnosis and QA-decision responsibilities cooperate through a controlled orchestration layer.
Eligible technical automation failures can be repaired and retried using confidence gates without changing the original business expectation simply to make a test pass.
Historical missions, failures and successful recoveries can help Auravis choose smarter regression scope and reuse validated knowledge.
CI/CD can eventually invoke Auravis and receive an auditable quality recommendation based on requirements, execution results and evidence.
The target is less repetitive manual QA execution and more human focus on product risk, requirement quality, governance and exceptional decisions.
Important boundary
Auravis is not intended to hide real defects or blindly make tests pass. AI can understand, plan and diagnose; Java controls state, policy and tool boundaries; Playwright performs permitted browser actions; evidence shows what actually happened. Sensitive actions and low-confidence recovery should remain governed.
Why this matters as an AI engineering project
The product demonstrates how a Java/Spring Boot system can combine requirement intelligence, RAG, AI agents, deterministic automation, PostgreSQL persistence, observability, Docker, CI/CD and AWS deployment around one meaningful end-to-end business problem rather than isolated AI demos.