Skip to content

00 — Context Summary: Current State Across All Repositories

Research date: 2026-04-15

Repository Landscape

AgenticAI-app (Laravel — Production)

Stack: Laravel 13 / PHP 8.4 / MySQL 8 / Redis 7 / Blade + Tailwind v4 + DaisyUI 5 / Docker + Kamal / GitHub Actions

What's built:

  • Scout (80 PHP files, production-ready) — End-to-end casual recruitment automation
  • 8 product categories (Forklift, Warehouse, Driver, Leadership, Food/Hospitality, Trades, Admin, General Labour)
  • Deterministic weighted scoring engine (critical 3x, core 2x, preferred 1x)
  • Three-track pipeline (Ready-to-Place, LSQ Review, Needs Onboarding)
  • Native assessment system (replaced Typeform)
  • Pool refresh from external databases (admin_central, admin_atslive)
  • Outreach (SendGrid email, SMSGlobal SMS)
  • Calendly interview scheduling
  • ASG Central onboarding integration

  • Billing/invoicing system

  • Verify (66 PHP files, production-ready) — AI-powered ID verification

  • 4 Claude AI agents (ClassifyDocument, ExtractFields, CompareFaces, CrossReference)
  • 5-stage pipeline: classify → extract → validate → face_compare → cross_reference
  • Provider failover: AWS Bedrock (primary) → Vertex AI Gemini (fallback)
  • AI budget management (per-org daily/weekly/monthly limits, circuit breaker)
  • Manual review escalation
  • Document retention with S3 deletion

  • OTP verification, demo mode

  • Reach — scaffolded, not built

  • Pulse — scaffolded, not built
  • Admin (13 files) — org management, product access, feature flags
  • Shared (16 files) — User, Organisation, ProductAccess, FeatureFlag, AuditLog
  • Orchestration (9 files) — cross-domain bridges (Scout ↔ Verify)

Database: 28 core tables + 2 external read-only databases

Testing: 1,220 tests (57 architecture + 109 unit + 1,054 feature), 3,722 assertions, 100% pass

Architecture enforcement: Deptrac (domain boundaries), PHPStan (Level max), Pint (PSR-12), PHPMD

AgenticAI-marketing (Astro — Production)

Stack: Astro 5 / Tailwind v4 / DaisyUI 5 / Cloudflare Pages

Product vision — four AI agents for labour hire:

  1. Scout — CV parsing, AI matching, automated outreach (68% screening time saved, 3.4x capacity)
  2. Verify — Right-to-work checks, licence validation, facial verification (6-second verification, 99%+ accuracy)
  3. Reach — SMS/voice/chat agents for worker and client communication (60%+ AI resolution rate)
  4. Pulse — Smart rostering, WHS compliance, timesheet processing, onboarding automation (57% HR admin saved)

Target market: Labour hire, warehouse, logistics, manufacturing, construction, food & beverage

Positioning: "The AI workforce for labour hire" — purpose-built, not generic HR software

AgenticAI-poc-elixir (Elixir — Completed PoC)

Stack: Phoenix 1.8.5 / Elixir 1.15 / PostgreSQL 17 / LiveView 1.1 / Oban 2.21 / Tailwind v4 / DaisyUI 5

Verdict: GO — Elixir better on 10/15 patterns, equivalent on 5, worse on 0.

What was built (4 domain contexts, ~50 schemas, 327 tests):

  1. Scout — Full recruitment domain (28 schemas). JobOrder state machine, ScoringService, ScoringPipeline, FeeCalculator, 4 Oban workers
  2. Verify — Full verification domain (24 schemas). 5-stage pipeline, PolicyEngine, BudgetLimitService, AI provider abstraction
  3. People — HR/payroll domain (12 schemas). Employee lifecycle state machine, PayrollPipeline (pure functions, Decimal precision, tested against ATO 2025-26 brackets), LeaveAccrual
  4. Shared — Cross-cutting (Organisation, MultiTenancy, Audit, BudgetGuard)
  5. Agents — AI agent orchestration. GenServer-based agents, DynamicSupervisor, AgentRegistry, Claude API client, Tool system (7+ tools), Intent router (pattern-matched + Claude fallback), LiveView chat UI

Key findings: - GenServer processes ARE agents (language-native, not bolted on) - Supervisor trees ARE self-healing fault isolation - Oban eliminates Redis (DB-backed jobs with PostgreSQL) - Pattern matching prevents entire bug categories in state machines - Pipe operator makes data pipelines natural - LiveView eliminates React/Vue entirely - Ecto's explicit queries prevent N+1 bugs - 30-40% less LOC than equivalent Laravel code

rnd_asg_central (CodeIgniter — Production Legacy)

Stack: CodeIgniter 3 / PHP 7+ / jQuery / MySQL / Amazon S3

Scale: 239 controllers, 203 models, 26,852 PHP files. Zero tests.

Features (comprehensive HR system):

  • Employee/candidate management with extensive data capture
  • Timesheets (individual, bulk, site-based, NFC clock-in, geo-location)
  • Rostering/scheduling (shift codes, rule sets, multi-offer, calendar view)
  • Leave management (request, approval, balances, unavailability)
  • Payroll processing (pay details, levels, allowances, superannuation)
  • Compliance tracking (functional assessments, crime checks, police checks, WWC)
  • Onboarding (multi-step profiles, document collection, inductions)
  • Client management (contacts, messaging, hierarchy)
  • Reporting (CSV, PDF, Excel export)
  • Communication (SMS, email, WhatsApp, Firebase push)
  • 14+ cron jobs (payroll, compliance, rostering, notifications)

Integrations: Multiple databases (admin_central, actatek clock system, admin_atslive ATS), OAuth (Google, Microsoft, Apple), S3, Firebase, DirectSMS, TransmitSMS, MessageBird

Technical debt: Massive controllers (100+ methods), 9,446-line general_helper.php, raw SQL queries, no tests, no documentation, plaintext credentials in config, multiple API versions, hardcoded client-specific code

rnd_asg_central_app (Flutter — Production)

Stack: Flutter/Dart (SDK 3.9.2-4.0.0) / BLoC / GoRouter / Dio / Firebase / NFC / QR

Features (field worker self-service): - Clock in/out via GPS, NFC tag, QR code - Roster viewing (weekly + calendar) - Leave requests and calendar - Time card management - Push notifications with action handling - Profile management - Background location streaming

23 screens, 40+ API endpoints

Not exposed on mobile: Payroll, expenses, performance reviews, training, team management, reporting, compliance admin, recruitment/onboarding

First-in-Last-out (Management — Confidential)

@Hexis / ASG Central v3 — Full scope:

Core modules (Phase 1): Recruitment, Onboarding, CMS (Employee Management), Rostering, Time & Attendance Additional modules (Phase 2): AMS (Asset Management), Mobile Apps (aCMS, aAMS, aTask, aTeam), Reporting & Analytics

Commercial plans:

  • Plan A (active): Phased build P1→P2→P3, $2.025M total. P1 $120K (mobile), P2 $315K (defence v2.5), P3 $1.59M (full platform)
  • Plan B (fallback if Plan A not started by July 1, 2026): Subscription-first model, $35K/month platform + $7.5K/month AgenticAI maintenance + variable AI credits

Client: Ashley Services Group (~5,000 casuals, 120% annual turnover, ~500 new starters/month)

Team restructuring planned: 15 offshore devs → 3 retained, triggered by P1 scoping sign-off

AgenticAI monetization: 4 products with per-function pricing. Full adoption estimated ~$27.7K/month credits + $7.5K maintenance

DISP-IRAP (Compliance)

Targeting: Entry Level DISP + Essential Eight Maturity Level 2

Gap analysis (2 April 2026): 10 main recommendations, target closure 14 May 2026

Implementation: 4 phases, 18 weeks, ~AUD 70K - Phase 1: Microsoft 365 E5 + Entra ID + MFA + Intune - Phase 2: Patch management + application control + PIM - Phase 3: Microsoft Sentinel SIEM + backups + incident response - Phase 4: Validation + audit readiness

Architecture: Zero Trust, identity as control plane, Microsoft-centric corporate security

Key Numbers

Metric Value
AgenticAI-app tests 1,220 (100% pass)
AgenticAI-app sprints 46+ completed
Elixir PoC tests 327 (100% pass)
Elixir PoC patterns validated 15/15
ASG Central v2 controllers 239
ASG Central v2 models 203
ASG Central v2 tests 0
Active casual employees ~5,000
Annual turnover ~120%
New starters/month ~500
Plan A total value $2,025,000
Current client monthly spend $41,000