Our work

Work we've built

Most client systems run inside their businesses and stay private, so we describe those without names. Martina Barakoska agreed to let us show her site.

Client project

Homepage of martinabarakoska.com, showing her name, a photo of her playing drums, and Listen now and Contact buttons.

Client website

Martina Barakoska: website and shop

Professional drummer, producer and educator

Her old site was hard to update, and her music, tour dates, shop and links sat on separate services.

What we built

  • A Shopify site that replaced her old WordPress site, with pages her team updates without a developer
  • A homepage built around her music, with streaming players, videos, testimonials and her social feed
  • Tour dates, interviews, gear and booking pages
  • A shop for her digital albums
  • A links page she controls, in place of a third-party link-in-bio service
  • A private press kit page kept out of search results
  • Search setup: page titles, descriptions, structured data and a clean sitemap, so her own site comes up for her name next to her brand and label profiles
  • Pages that load fast on a phone, which search engines weigh

Built with: Shopify, with custom page sections and templates

Visit the live site

AI and automation

AI and automation we've built

In each of these, AI drafts or proposes and a person approves. Client names are removed. Three include a demo you can try.

  • AI and automation

    AI agent control room with human approval

    Built for our own team

    With several AI workers running across projects, we couldn't see what they changed or stop a bad change before it shipped.

    What it does
    • One dashboard for agents, tasks and activity
    • An approval queue that records who approved or rejected each change, and why
    • Audit logs, alerts and a summary of what happened while you were away
    • Small changes only, with a checkpoint before each one so we can roll back
    • The system pauses itself after repeated failures

    Built with: A web app with a local database, scheduled jobs and several AI models

    Try it

    Approval queue
    • Update the weekly summary layout
    • Retry a failed sync after a short wait
    • Rename a column in the task view

    Applied

    • Nothing applied yet

    Nothing an agent proposes goes live until a person approves it.

  • AI and automation

    Email campaign drafting with an approval step

    A software products company

    The team wrote each marketing email from scratch with no plan and no record of past sends.

    What it does
    • An indexed archive of past campaign emails
    • A calendar of planned emails for the coming weeks
    • Drafts in the brand's voice, based on the plan
    • An approval step that turns a draft into a ready email
    • The tool can't send: a person sends each email

    Built with: A drafting tool, an email archive and an approval step that feeds the existing email platform

    Try it

    1. Week 1New feature tips
    2. Week 2Customer story
    3. Week 3Seasonal update

    Drafts

    No drafts yet

    Ready to send

    Nothing approved yet

    The tool drafts from the plan. A person approves, and a person sends.

  • AI and automation

    Shared memory for AI assistants

    Built for our own team

    Our AI assistants forgot decisions between sessions and couldn't see what the others had learned, so we explained the same facts over and over.

    What it does
    • One memory store that several AI tools read and write at once
    • Keeps decisions, preferences and project facts, and skips duplicates
    • Loads recent context at the start of each session
    • Search by meaning, plus a deeper search for harder questions
    • Keeps corrections next to the facts they fix

    Built with: A local database with meaning-based search and a standard AI tool connector

    Try it

    Assistant A

    The Friday report goes to the whole team.

    Assistant B

    Who gets the Friday report?

    Ask to see what B knows.

    Save a fact with one assistant, then ask the other.

  • AI and automation

    Private AI assistant for one executive

    An executive client

    The client wanted an AI assistant they could message from their phone, without sharing an account or knowing anything technical.

    What it does
    • The client chats with it in a messaging app they already use
    • One app, licensed to the client's own computer
    • Runs on older and newer Mac processors
    • A license service so we can manage access remotely
    • A ten-minute setup guide that walks a non-technical person through each step, including the few Terminal commands

    Built with: A desktop app, a chat connection, a small cloud license check and an AI model

  • AI and automation

    Video production pipeline for a kids' learning channel

    A children's education brand

    Each learning video needed several versions, thumbnails, captions and upload details. A small team couldn't keep up by hand.

    What it does
    • Renders animated episodes from templates in vertical and widescreen formats
    • Makes alternate versions of each lesson from one template
    • Builds thumbnails and captions in batches
    • Schedules uploads and holds each public video until someone approves it

    Built with: A code-based video renderer, batch scripts and the video platform's upload tools

Get in touch

Start a project

Every project starts with an intro meeting, so we understand your business before we scope the work.