A suite of web apps, automations, and AI tooling I designed and built to run operations, logistics, quality, and product support at RadiusXR - a medical-device company shipping to a fleet of roughly 4,000 devices. Most are Google Apps Script web apps backed by Google Sheets; a few are Node.js services and one Electron desktop app.
These tools were built for internal use, so specific integrations, data flows, and access details are intentionally omitted. What follows is a high-level tour of what each one does and the stack behind it. Contact me for a private walkthrough.
Kitting and production-tracking system that provisions device tablets, tracks component traceability, and generates build records. Mobile and desktop surfaces let the floor scan units and provision a full system in under 5 minutes, down from 30 to 45 minutes of manual entry and hand-assembled PDFs. A multi-tab Sheets data model with batched writes and row-level locking keeps concurrent operators consistent.
Tiered rollout framework for pushing app and configuration updates across the ~4,000-tablet fleet. Tier-based rollouts, dynamic-group fallback for legacy devices, and automatic OAuth token rotation brought the rollout failure rate down from about 30% to under 5%.
Real-time analytics surface that folds product usage, trial conversion, and support metrics into one leadership view. Behavioral data from Mixpanel is cached for speed, written summaries are generated on demand through the in-house Claude gateway, and trial-age logic flags accounts nearing conversion deadlines.
Release user-acceptance-testing tool that logs results, drafts Jira issues with Gemini, and collects stakeholder sign-off. Release-scoped status tracking gates every fleet release before it ships to the field.
Return-merchandise-authorization tracker that coordinates inbound shipments, customer records, and return status across support and logistics. It syncs Salesforce return records and drives a refurbish-to-return-to-stock flow, reconciling inventory as units move through inspection and back to the shelf.
Automated carrier-tracking pipeline that collects shipment status on a schedule and writes it into shared operational spreadsheets, giving ops a live view without manual carrier lookups.
Shared LLM gateway that lets any internal tool submit a prompt and poll for a result. A sheet-backed job queue is processed by a Node worker running Claude, with model routing, streaming tool-use progress, and scoped access. Reachable by every internal app and handling 50 to 100 prompts a day, it replaced scattered direct API calls with one governed entry point.
Desktop operations hub that aggregates outstanding tasks from Gmail, Slack, Notion, Jira, and Calendar through scheduled AI routines, surfacing one prioritized action list each morning.
Automated customer-onboarding flow that provisions accounts from parameterized Salesforce links. It cut onboarding support calls by 80% and supported a major distribution partnership.
Campaign tool for the support team that assembles targeted, personalized email drafts for human review before send, keeping a person in the loop on every outbound message.
Product-management workspace: a hash-routed single-page app for roadmap, feature status, and structured intake of meeting notes into trackable items.
Production-floor suite covering device programming, a phone-based floor-scan scanner, and an event-sourced unit-master inventory that other tools query over a JSON API. Activity logging and archive-on-swap keep device history intact as components change.
All descriptions are high-level and non-sensitive. Architecture, endpoints, credentials, and sheet structures are not public.