boringstack

End-to-End Back-Office AI Architecture

05
Client
Interface
04
Agent
Execution
03
Intelligence
(The Moat)
02
Context
Engineering
01
Governance &
Observability
CLIENT INTERFACE
boringstack Portal
Dashboards • Exception Queue
Audit Log • Workflow Monitor
Slack / Teams
One-click approvals • Alerts
Daily/weekly digests
AGENT EXECUTION
Workflow Agents
Single agent per domain
Scales to multi-agent
Tools + Skills orchestration
Trigger Engine
Scheduled • Event-driven
API • Human-initiated
Approval Router
Tiered autonomy:
Auto • One-click • Full review
INTELLIGENCE (THE MOAT)
Workflow Map
Live ontology: business objects,
relationships, governed actions
Auto-inferred + FDE-curated
Decision Graph
Decision commits + outcomes
Pattern extraction (cross-client)
Reinforcement learning loop
Skill Library
Templates per function
20-30% customisation/client
Compounds with every deploy
CONTEXT ENGINEERING
MCP Connectors
FDE-built per client
PII / non-PII scoped
Role-based access control
Data Classifier
PII / Sensitive / Public
Mapped during diagnostic
Enforced at connector level
LLM Gateway
Bedrock / Azure OpenAI
Model-agnostic routing
Data residency controls
GOVERNANCE & OBSERVABILITY
Audit Trail
Every action logged + traced
Immutable per-client log
Exportable for external audit
Agent Evals
Dry-run mode before production
Accuracy tracking vs baselines
Agents earn autonomy over time
Environments
Sandbox + Production
Simple promotion process
FDE validates, ops approves
PII Systems of Record
Xero / MYOBBambooHRDeputyPayrollCRMBankingHRIS
Non-PII Systems of Record
Slack / TeamsConfluenceNotionGoogle WorkspaceProject ToolsShared Drives
Ops Managers
3-5 people, daily use
Review exceptions + approve
Finance, HR, Compliance leads
Leadership / PE
Dashboards, quarterly
EBITDA + automation metrics
Cost displacement tracking
Forward-Deployed
Engineers
Build + maintain agents
boringstack team
Claude Code + Skills