Back to Projects

Aviation Tool Inventory — Closed-Loop Accountability with AI Trend Monitoring

A purpose-built, closed-loop digital tool inventory system with an AI-powered trend monitoring layer — designed specifically for aviation maintenance organizations under Transport Canada CARs Part 145 and FAA Part 145. The operational core runs within the organization's own AWS VPC: formal tool lifecycle state machine, personnel checkout with aircraft tail number assignment, admin notification on every status change, and barcode-reconciled audits. The AI trend agent operates on aggregated event summaries only and delivers plain-language reorder and attrition recommendations to the admin dashboard. The system is designed to pay for itself within the first grounding event it prevents.

Core Technologies

PostgreSQLNode.jsReactAWS (RDS / EC2 / VPC)Claude APIREST APIBarcode Scanning

Architecture Components

  • Closed operational database within a private AWS VPC — all tool records, transitions, and calibration data fully within the organization's control
  • Formal state machine enforcing aviation tool lifecycle: Serviceable → On Loan → WMTC (broken) → Calibration Centre → Serviceable — illegal transitions rejected at the system level
  • Personnel ID checkout with mandatory location assignment: hangar bay, workbench, or aircraft tail number — every loan is a named, located, timestamped record
  • Admin notification system — immediate escalation on every status change and missed calibration due date
  • Barcode scan handler for rapid check-in/check-out and periodic audit reconciliation against live inventory
  • AI trend agent (Claude API) operating on aggregated, anonymized event summaries — no raw personnel data transmitted externally
  • Four-signal trend engine: systematic low stock, loss/attrition patterns, overuse/underuse anomalies, and breakage frequency with supplier vs. usage attribution
  • React admin dashboard: live colour-coded status board, per-aircraft accountability reports, calibration compliance calendar, and exportable AI insight panel
THREE-LAYER CLOSED-LOOP ARCHITECTURELAYER 1OperationalCorePrivate AWS VPCPostgreSQLRDS — tool recordsState Machinelifecycle enforcementNotificationevery transition → adminBarcode APIscan → checkout/returnAudit Loggeractor + timestampAll raw operational data stays within the VPC. No personnel records leave this boundary.AGGREGATION BOUNDARY — summaries only cross hereLAYER 2AI TrendAgentEvent Aggregator30-day rolling summaryClaude APIclaude-sonnet-4-64-Signal Trend Analysislow stock · loss/attrition · overuse/underuse · breakage frequencyOperates on aggregated event summaries only. No individual personnel data, locations, or raw timestamps.LAYER 3AdminDashboardLive Status Boardcolour-coded tool statesPer-Aircraft Reportstools on loan by tail numberCalibrationCalendardue dates + overdue flagsAI Insight Panelrecommendations + exportAudit trail always current. Compliance evidence on demand — never reconstructed before an inspection.
TOOL LIFECYCLE — FORMAL STATE MACHINETransport Canada CARs Part 145 / FAA Part 145 — all transitions logged with actor, timestamp, locationSERVICEABLEavailable for useON LOANID + location assignedWMTCbroken — quarantinedillegal transitions blockedCALIBRATIONCENTREout for recalibrationcheckout — personnel ID + location requiredreturn — pre-check completefault reported⚡ admin notifiedneeds recalibration⚡ admin notifiedcalibratedirreparableCONDEMNEDterminal — removed from service permanentlysolid arrow = valid transition dashed = return path ⚡ = admin escalation triggered

The Cost of Manual Tool Accountability in Regulated Aviation

Aviation maintenance organizations operate under one of the strictest accountability frameworks in any industry — yet most still manage tool inventory through paper logs, tag boards, and spreadsheets. The gap between regulatory expectation and operational reality creates compounding risk: FOD incidents, failed audits, grounded aircraft, and technician hours bled into searching for tools rather than turning wrenches.

  • FOD incidents caused by untracked tools cost the global aviation industry an estimated $14.5 billion per year; a single FOD engine damage event can exceed $1 million in repair costs before accounting for grounding time, crew disruption, and regulatory investigation
  • Manual operations lose $36,500–$120,000 annually at a mid-size maintenance organization (10–20 technicians, 500–1,500 tools): search time, audit prep scrambles, calibration overruns, and reactive purchasing — all preventable with real-time digital accountability
  • Off-the-shelf asset tracking platforms are built for general inventory, not aviation: no formal status state machine, no per-aircraft tail number assignment, no regulatory-grade audit trail, and no AI trend intelligence layer

A Three-Layer Closed-Loop Architecture

A purpose-built system aligned to the actual workflow of an aviation tool crib — closed at the operational core for data sovereignty, AI-powered at the trend layer for predictive intelligence, and clean at the surface for administrative visibility and compliance.

  • Layer 1 — Operational core: formal state machine (Serviceable → On Loan → WMTC → Calibration Centre), named/located/timestamped checkout records, barcode audit reconciliation, admin notification on every status change
  • Layer 2 — AI trend agent: Claude API integration on aggregated event summaries only; four signal types (systematic low stock, loss/attrition, overuse/underuse, breakage frequency) delivering plain-language recommendations to the dashboard
  • Layer 3 — Admin dashboard: live colour-coded status board, per-aircraft accountability reports, calibration compliance calendar, and exportable AI insight panel for procurement decisions
  • Regulatory alignment built in from the foundation: every status transition logged with actor, timestamp, and location — audit trail always current under Transport Canada CARs Part 145 and FAA Part 145, never reconstructed before an inspection

Return on Investment

Conservative Year 1 savings exceed infrastructure costs by more than 20x. Long-term value compounds as the AI trend layer accumulates pattern data. The system pays for its entire build cost the first time it prevents one aircraft from sitting on the ground for a day.

  • Year 1 savings (conservative): $18,500 from search time elimination, reduced audit prep, calibration compliance, and smarter reorder decisions — on $38–$68/month in AWS infrastructure
  • Year 1 savings (realistic): $42,000 — infrastructure ROI exceeds 50x at this level, with savings realized within the first 90 days of full deployment
  • Long-term annual benefit (Years 2–5): $16,000–$43,000/year from AI-driven reorder optimization, reduced tool loss and attrition, and on-demand audit readiness as the trend agent accumulates pattern data
  • FOD prevention value: a single avoided grounding event ($3,000–$15,000 direct cost) exceeds the full annual infrastructure cost; an avoided FOD damage event ($50,000–$1M+) covers the full system build cost

Key Learnings & Decisions

Built for Aviation, Not Adapted to It

  • The formal status state machine — Serviceable → On Loan → WMTC → Calibration Centre — is not a feature bolted onto a general platform. It is the foundation. Aviation tool accountability has a specific workflow; the system is designed around that workflow, not the other way around.
  • Per-aircraft tail number assignment is the key accountability primitive that general asset tracking platforms miss. A tool on loan must be locatable: hangar bay, workbench, or specific aircraft. Without this, the system cannot answer the pre-return check question that actually prevents FOD.
  • Admin notification on every status change is non-negotiable in a regulated environment — it is the compliance mechanism, not a convenience feature. The system's value proposition depends on it firing reliably on every transition.

Closed by Design, Open Development Surface

  • The operational core is closed by default: deployed within the organization's own AWS VPC, with no raw operational data transmitted externally. This matters for organizations handling sensitive aircraft registrations, personnel records, and operational schedules.
  • The AI trend agent is deliberately designed as the open development surface. The trend logic, prompt engineering, and recommendation models can be iterated independently of the operational core — as the organization's needs evolve, the intelligence layer evolves with them.
  • The aggregation boundary between operational data and AI analysis is an architectural constraint, not just a privacy preference. It ensures the trend agent never receives individual personnel data, which simplifies both the privacy posture and the prompt design.

Regulatory Alignment

  • Transport Canada CARs Part 145 and FAA Part 145 are the compliance baseline, not an afterthought. Every status transition is logged with actor, timestamp, and location. The audit trail is always current — it does not need to be reconstructed before an inspection.
  • The state machine's illegal transition enforcement serves a dual purpose: it prevents operational errors (a condemned tool cannot be accidentally returned to service) and it creates a tamper-evident transition log where gaps or reversals are immediately visible.
  • Calibration due date tracking is a regulatory obligation, not a nice-to-have. The notification system treats missed calibration dates as escalations — the administrator is notified immediately, not at the next manual review cycle.

Implementation Milestones

A breakdown of the key tasks and milestones that brought this project to life.

Phase 1 — Operational Core

Planned

Database schema, formal state machine, REST API, barcode scan handler, admin notification system, role-based access control, and cloud deployment on AWS within a private VPC. Estimated delivery: 8–12 weeks. At completion: a fully functional digital tool crib with real-time accountability and a complete, always-current audit trail.

Phase 2 — Admin Dashboard

Planned

React-based frontend: live colour-coded status board, loan management interface, calibration compliance calendar, per-aircraft accountability reports, and audit log viewer. Estimated delivery: 4–6 weeks following Phase 1. At completion: paper logs and tag boards are fully replaced.

Phase 3 — AI Trend Agent

Planned

Event aggregation pipeline, Claude API integration, four-signal trend analysis (low stock, attrition, overuse/underuse, breakage frequency), and recommendation display in the admin dashboard. Estimated delivery: 4–6 weeks following Phase 2. At completion: the organization moves from reactive tool management to predictive inventory intelligence.

AI trend agent — aggregated event summaries routed to Claude; no raw personnel data crosses this boundary

Loading code...