SENTINEL TOPOGRAPHY

SYSTEM_AUTONOMIC: ACTIVE

ARCHITECTURAL DEEP-DIVE v1.0.0 (GKE EDITION)

The Living Engine

A biological approach to software quality. While currently operating autonomously via local Docker, Sentinel's Future Vision evolves it into a distributed organism living on GKE Autopilot, creating an autonomous nervous system for your applications.

LAUNCH_MISSION_CONTROL

The Autonomic Fix Loop

How Sentinel transforms a user's natural language into a self-healing Java execution.

User / Recruiter

Visits Portfolio

Portfolio Portal

Triggers Sentinel Demo

Sentinel Core

Mission Control UI

Cortex AI

Gemini 2.0 Decisions

GKE Cluster

Batch Jobs & Selenium

Data Backbone

GCS Fuse & Firestore

Future Architecture (WIP)

The "Split-Brain" Microservice Architecture

While Sentinel currently operates effectively via Docker locally, this represents our Work in Progress (Future Implementation) design. The goal is decoupling Intelligence from Execution for infinite scalability on GKE Autopilot.

GKE Autopilot Cluster

us-central1

WORK IN PROGRESS: FUTURE IMPLEMENTATION

NODE_POOL: STANDARD

Sentinel Web

Next.js Control Plane

Ram: 4GB

CPU: 2v

GCS FUSE

Read/Write Shared Volume

NODE_POOL: SPOT

Runner Jobs

Ephemeral Workers

-60% Cost

Auto-Retry

Matrix Logs (SSE)

We replaced database polling with a direct Server-Sent Events (SSE) pipe from the Kubernetes Pod.

kubectl logs -f -l missionId=123 | stream

Kubernetes Jobs

Runners are now Batch Jobs, not raw Pods. This gives us "Spot Preemption Protection" — if Google kills a node, the Job auto-restarts elsewhere.

backoffLimit: 2
ttlSecondsAfterFinished: 60

Warm Pool Strategy

A single "Standby" pod keeps the 2GB Docker image pre-cached on the cluster nodes.

Cold Start: 3m 00sWarm Start: 0m 05s

The Biological Brain

Cortex v1.0 Learning Loop

A real-time visualization of how Cortex detects failure, analyzes the DOM via Gemini, and injects self-healing code.

cortex-daemon — zsh — 80x24

$ sentinel run --mission=e-commerce

DETECTED_FAILURE

Element 'SubmitButton' not found

CORTEX_ANALYZING

Processing DOM Snapshot & Accessibility Tree...

GENERATING_FIX

Suggestion: driver.findElement(By.id('submit-btn'))

HEALING_INJECTED

Re-running test step... SUCCESS

➜

Sentinel Architecture

The Biological Model (v1.0)

SELF-HEALING 1.0

The EyeVision

Deep Spy

Sees the 'Deep Reality' (Shadow DOMs) and detects visual obstructions code cannot perceive.

Implemented via: DOMCrawler.js + Screenshots

The BrainCognition

Triangulation Engine

Decodes 'User Intent' vs 'Reality'. Understands the semantic role of locators beyond IDs.

Implemented via: Gemini 2.0 Flash + System Prompt

The SenseProprioception

Sentinel Wrapper

'Feels' application resistance. Automatically shifts strategies when elements are occluded.

Implemented via: Aggressive Waiting / JS Directives

The MemoryExperience

Semantic Cache

Reacts with muscle memory. Instantly repeats successful historical fixes for known patterns.

Implemented via: Firestore + Hash Triangulation

Fix Loop Triangulation

APP DOM

CORTEX fixed

JAVA OBJ

The Guardrail System

Autonomous Triage & Self-Correction

In v1.0, we introduced the Defect Tracker with Headless Probes. This system eliminates the "Manual Re-test" phase by delegating verification to autonomous workers.

Defect "Hotness" Indicator

A real-time analytics module that prioritizes failures based on their "frequency bleed." If a failure is detected across multiple features or repeats over time, the left border glows intense red, elevating it to CRITICAL status automatically.

Headless Validation Probes

When an engineer (or AI) applies a fix, they trigger a "Probe." Sentinel spawns an ephemeral, headless GKE Job that runs *only* the affected feature. On success, it communicates via the `/api/cortex/probe` endpoint to **auto-close** the ticket.

PROBE_LIFECYCLE_LOG

[PROBE] Spawning headless job probe-991f...

mvn test -Dcucumber.filter.tags=@social_links -Dheadless=true

> Scanned 1 feature...
> Executed 3 scenarios...
> BUILD SUCCESS (12.4s)

[CORTEX] POST /api/cortex/probe { status: 'Passed' }

[SYSTEM] DEFECT #991f STATUS -> CLOSED

LIVE UPGRADE DEPLOYED

Cortex Defect Tracker v1.0.0

Auto-Close Logic

FIXED: Tests that pass now automatically resolve their corresponding open defects.

Stale Probe Timeout

FIXED: Safety net resets any "PROBING" status older than 10 minutes, preventing limbo states.

Source Code Viewer

FEATURE: Secure modal showing live source code (.java files) directly in the dashboard.

"The dashboard now provides a true real-time view of your project's health."

Knowledge Synthesis

Collaborative Quality Hub

Closing the feedback loop between the AI's autonomous decisions and the QA team's tribal knowledge.

120+

AI Fixes

Threads

Failure Context Threads

Every automated failure is automatically tagged with a discussion thread. Engineers can upload screenshots or link legacy reports, giving the AI better context for next time.

Semantic Attachment Bridge

The `CommunityHub` uses Firebase Admin Storage to handle large artifacts (Videos, HTML Reports) and associates them semantically with specific test suites.

Knowledge Distillation

Cortex "listens" to the human conversations in the Hub. If a human says "This is a known issue with the Staging DB," Cortex will stop trying to "heal" the code and flag it as an Environment Issue.

Technical Repository Architecture

High-level mapping of the Sentinel & Cortex ecosystem.

PROTECTED INTELLECTUAL PROPERTY

ProjectDriverManager.java

Project Explorer

E-Automate

src

main

java

test

java

sentinel-core

DriverManager.java×

1// [PROPRIETARY SOURCE CODE]
2// Dynamic Grid Orchestration & Browser Lifecycle Management
3// Content redacted for intellectual property protection.
4// 
5// Features:
6// - Multi-Protocol Proxy Tunneling
7// - Atomic State Synchronization
8// - Automated Resource scavenging

e-automatesrc...DriverManager.java

The Growth Curve

The Mock Brain

Initial regex-based engine to validate the Gherkin parsing pipeline without LLM costs. Established the 'Sentinel' API route.

LLM Integration

Integrated Gemini 2.0 Flash. Transformed static regex logic into dynamic Intent Analysis that could understand natural language directions.

Robust Isolation

Separated AI execution from the Regression Suite using strict tagging conventions (@ai_generated) and safety flags.

Containerization

Moved execution from Shell to Docker. Solved cross-container networking issues using 'host.docker.internal' to allow the isolated runner to test the localhost Next.js app.

Self-Healing Architecture

The 'SDET' module. Created the `CodeInjector` and `GherkinAnalyzer` which detect missing steps and write Java code on the fly.

Targeted Search (React Fix)

Solved complex synchronization issues with React's Virtual DOM by implementing atomic char-by-char typing in the AI steps.

Real-Time Observability

Built the 'Mission Control' dashboard with live Firestore synchronization. Integrated VNC-over-Web for visual validation of AI runs.

Autonomous Triage

Implemented the Cortex Defect Tracker with automated headless probes that self-resolve defects in Firestore upon verification.

Collaborative Quality

Launched the Community Hub and Failure Discussions, bridging the gap between AI automation and human tribal knowledge.