John EdrickNubla.

What I built

• An MCP server published to npm that gives Claude Code eight tools for reading and updating tickets, scoped per project by token
• One-click coding agent: a ticket becomes an isolated git worktree, a headless Claude Code run, and a pull request — nothing ever auto-merges
• A local agent daemon that leases queued runs and keeps source on the developer’s machine; only status, diff stats, and usage are reported home
• Pilot orchestration agents that batch many tickets into a single Claude session, review their own branches, run a fix pass, and open pull requests — drivable from a phone
• Kanban project boards with drag-and-drop lanes defined per project, gated so a ticket cannot reach “ready for review” without a pushed branch or an open pull request
• An in-app AI assistant that answers questions against live tickets and sprints, and turns a conversation into one or many tickets behind a confirmation step
• Lead campaigns with configurable pipeline stages, so prospects are tracked in the same workspace as delivery
• Database-per-tenant isolation with plan features, module toggles, and role permissions stacked as three independent gates
• Invoices generated from selected time logs — minutes summed across breaks and midnight, Dutch VAT applied, PDF emailed
• An Outlook add-in that turns an email into a ticket, proposing project, priority, and story points in a single LLM call

What I built

• AI photo studio: uploads go to Google Gemini, which replaces the background with a consistent studio wall, floor, and lighting
• A single instruction handles three cases — non-car images are detected and returned untouched, and interior shots are repainted only through the window glass
• A branded plate is rendered onto the car in correct perspective, so every listing photo brands itself without a compositing step
• Originals are always preserved beside the generated version, so an operator can regenerate with an extra instruction and never compound artifacts
• Batch uploads processed on a queue, with per-image progress pushed live over WebSockets and listings auto-publishing once the last photo lands
• Purchase records carrying import costs, BPM, and Dutch/German VAT status per vehicle, feeding margin calculations
• Invoicing and credit notes across two companies, reconciled against bank transactions via GoCardless and ING PSD2
• Chauffeur route planning with map-based stops and generated pickup, delivery, and checklist paperwork
• Saved views over the same inventory — compact, financial, VAT, paperwork, planning, pickup, delivery, debtors — each with its own columns and deep filtering
• Records edited in place straight from those tables, alongside personal, closed, and delivery-planning dashboards charting bought and sold volume against targets

What I built

• A booking engine that validates every appointment against working hours, availability blocks, and existing bookings inside one transaction
• Multi-service appointments with price and duration snapshotted onto the booking, so history stays accurate when the catalog changes
• Public per-shop booking pages — rate-limited and reCAPTCHA-scored — that enforce which barber may perform which service
• A walk-in queue running alongside scheduled appointments, with the queued state restricted to walk-ins at the domain layer
• Multi-branch shops with Google Places geocoding, per-branch opening hours, and configurable slot intervals
• Google Sign-In with server-side ID token verification written by hand: signature, issuer, audience, expiry, and verified email
• Dashboards for daily and monthly revenue, top services, and staff performance, built on dedicated repository queries
• A Flutter staff app for iOS and Android sharing the same API and role model as the web workspace

What I built

• Event setup in a few clicks, each night getting its own shareable link
• A unique QR code per event, so guests join by pointing a phone camera at it
• Song search against the Spotify catalogue, so every request resolves to a real track instead of a half-remembered title
• Requests arriving in the DJ’s queue through the night, so the booth can read the room at a glance
• A request page built mobile-first, because every guest arrives on a phone in a dark room
• DJ accounts with a dashboard holding past and upcoming events