Parts Intelligence

At a glance

• Outcome: Reconciled 30,000+ assets across 2,300+ models into dependable part identities. Collapsed duplicates and attached confidence levels to compatibility.
• Status: Working private tool. Another engineer’s internal rebuild retained the reverse engineering and deduplication; compatibility and identification stand alone.
• Role: Built solo alongside my repair role.
• Stack & libraries: Python, SQL, nightly ingestion and an on-device vision model.
• Source: Private employer tooling with vendor and internal names generalized. This page is the case study.
• Validation: Checked reconciliation against known-good records and traced linked assets across shipment and receipt tables. Techs used the identifier in the field.
• Limitations: The identifier is barebones. Confidence scoring is per-source rather than per-claim, so a well-sourced wrong fact still scores well.

Components

• Loaded nightly exports from a legacy medical-device inventory app without a real API into a queryable store. Reconciled two disagreeing ID systems and canonicalized duplicate and near-duplicate assets.
• Combined shipment records, service-manual PDFs, email archives, supplier and parts-shop listings, and firmware reverse engineering in a fit-and-substitution store.
• Identified photographed part labels with an on-device vision model and added the result to the store.

Repair Shop Automation documents the related daily workflow tools.