Luke McCarter · Toronto

philomath is...

...a lover of learning. φíλos μανθανειν
Over twenty years in industrial maintenance operations, environmental compliance, and continuous improvement, driven by emerging technology and learning something new every day.

Short version

Twenty years building maintenance operations for industrial processing and manufacturing. 3,000+ assets across multiple sites. My professional life is CMMS implementation, predictive maintenance, environmental compliance, and building data pipelines to help make the invisible parts of operations visible.

The creative

Lifelong violinist and drummer, playing in Oakville Symphony and indie-rock bands. Listening habits run from Mozart to Meshuggah, Grappelli to Gojira, Brahms to Beastie Boys. Stipple artist and VW tuning enthusiast. Custom watchmaker creating full builds from sourced movements, cases, and hand-painted dials. Trained audio engineer with a project studio. The thread connecting all of it is curiosity; how things work, and how far the tools within reach can take you.

The technical

Self-taught technologist since my first DOS command on a Commodore 64. Built Geocities sites in HTML before Google existed. Decades of hands-on experience across personal, business, and creative software categories. Now building AI-assisted tools, industrial IoT pipelines, and workflow automation for heavy industry. Still just as curious about what the next command does. Active member of the Association for the Advancement of Artificial Intelligence (AAAI).

Things I'm working on

Diagnostic Assistant

In Progress
First-run onboarding — user guide walkthrough and diagnostic setup AI diagnostic in action — photo submitted, visual guide with manual reference returned Resources — manuals, troubleshooting guides, remediation history with verified resolutions

An AI diagnostic coach that walks maintenance technicians through fault identification and repair on niche industrial equipment. Built around manufacturer service manuals for an industrial wire strapping system; the kind of machine where institutional knowledge lives in one person's head until they retire. The interface guides rather than guesses: asking for specific photos mid-conversation, presenting conditional decision trees, and citing exact manual references.

RAG over manufacturer service manuals · Multi-turn conversational UI · In-context image analysis · Guided decision trees with conditional branching · Per-equipment knowledge base

Priority Pulse

Open Source
Pipeline diagram
Priority Pulse v1.6 · Daily Triage · 2026-03-08 (Saturday) · Sources: 9/9 active
▸ Quick Stats
Active Time7h 12mSwitches412FRAGModeReactive 68%Deep Work2 blocksBest Focus38m (CMMS)Fragmentation94/100
▸ Time Allocation
Email / Comms
2h 34mCOMMS CMMS (UpKeep)
1h 07mDRIFT Vendor Coord.
52m Fleet / MTO
41m Environmental
28m Team / Sched.
35m Admin / Other
55m
▸ Triage
Respond to conveyor pricing requestVP Ops waiting 10 days. Apex Industrial quoted $14,200 — needs sign-off.
Process 4 overdue Google Tasks"Update UpKeep PM templates," "Confirm delivery schedule," "Submit Q1 safety report," "Send PO-4481"
Block 2-hour CMMS windowUpKeep PM schedule development needs protected time. Two short bursts instead of deep work.
ClearWater SWM inspection prepThursday inspection confirmed. Weather station API linked, pre-inspection checklist not started.
Apex Industrial compressor belt installAwaiting OEM belt dimensions (6 days, no response). Ref: WO-7823.
▸ Personal Support
Email as Default Mode — 35.7% email time correlates with unplanned gaps. Pre-load a "default task" for gaps to redirect energy toward CMMS.
CMMS Avoidance Signal — Two short sessions (38m, 29m) suggests task initiation friction. Open UpKeep first thing before Outlook.
Fragmentation is Structural — 58 switches/hr is driven by the role. Protect the two daily deep-work blocks rather than reducing switches.

A local-first productivity intelligence system built for neurodivergent brains in high-interrupt operational roles. Passively collects data from nine sources throughout the workday: UI activity, email, calendar, meeting transcripts, phone calls, browser history; packages it into a consolidated daily analysis file, and delivers AI-triaged feedback via the Claude API. Surfaces dropped tasks, flags strategic drift, and provides honest time-allocation analysis. No cloud, no telemetry.

Python · Claude API (Haiku for classification, Sonnet for analysis) · OAuth 2.0 multi-source ingestion · Power Automate pipeline · Obsidian vault output · Local-first, no cloud

Scrap Metal ID

In Progress
Capture interface — camera or upload for on-the-spot classification AI prediction with expert verification — correct or reclassify to retrain
ML dashboard — 94% accuracy across multiple material categories

A mobile classification tool that helps yard workers identify scrap metal grades on the spot. Trained on 15,000+ images using Nyckel, achieving 98–99% accuracy on broad material categories and 78%+ on subcategories and contamination detection. The model improves continuously because the people using it are the ones training it; expert corrections on the floor feed directly back into the training pipeline.

Custom ML image classification (Nyckel) · On-device inference · Continuous learning from expert corrections · Workflow-integrated capture/classify/confirm UX · Ferrous + non-ferrous multi-grade taxonomy

Part Lokker

In Progress
Part Lokker — document extraction to editable parts table CMMS parts inventory — destination for extracted data

Drop a photo of a quote, invoice, or packing slip and get a clean, ready-to-import file for your CMMS. Eliminates the manual data entry bottleneck in parts management, allowing technicians to photograph a vendor document on the shop floor and the system extracts part numbers, descriptions, unit costs, and vendor details into a formatted spreadsheet that maps directly to the CMMS import specification.

Claude API document extraction (OCR + LLM) · Multi-format input (.pdf, .doc, .jpeg, .png) · Auto-formatted .csv to CMMS import spec · Field mapping for vendor-specific formats · Batch processing

VRespite

Parked

VR applications with two focus areas: rapid-access sensory environments for students experiencing autism-related sensory overloads at school, and custom-built immersive experiences for long-term and palliative care patients. Validated ROI thesis with Peel Region special education teachers and a Toronto District School Board senior manager. Three-person founding team: prototyping and production (Luke), business development (Sr. Director, Canadian division of a publicly traded company), and operations/legal (M.B.A., Chief of Staff, global IT company).

360° video production · Purpose-built VR environments · Clinical + educational deployment · Sensory modulation research-informed
Stipple portrait — Bernese mountain dog Custom diver watch — hand-painted blue dial Stipple drawing — small bird Ink drawing — Nissan Skyline Custom watch — textured dial with orange accents Project studio — drum kit MK7 GTI — rear quarter Stipple drawing — bass guitar Stipple portrait — Bernese mountain dog face MK7 GTI — engine bay

Say hello