About / A workbench with a story

Built from
a real need.
Made to grow.

Better models begin with better understanding. BadBob brings the work around computer vision into one place, so you can see what to improve and take the next step.

Follow the story
BadBob's robot exploring a library full of books
A curious mind.A practical purpose.
01 / How it startedOne question led to the next

From a dice game
to a vision workbench.

BadBob grew from using the tools, noticing the gaps and building a clearer way through them.

  1. 01

    A real-world starting point

    DiceScore

    It started with DiceScore: a dice game, a camera and a model learning to recognise what was in front of it. Training a model raised another question: how do you understand where it succeeds, where it struggles and what to improve next?

  2. 02

    Learning from the results

    Failure Bank

    The first answer was a Python application called Failure Bank, built to inspect the results of Create ML training. It made the mistakes worth looking at, so they could inform the next round of work.

  3. 03

    From understanding to correcting

    Annotator

    Reviewing results led naturally to working on the data itself. Annotator brought image annotation into the workflow: a place to draw, correct and refine the examples a model would learn from.

  4. 04

    Bringing the workflow together

    BadBob

    Those connected needs grew into BadBob, a native macOS workbench for capture, annotation, dataset preparation, training and evaluation. Dice was the beginning; the workbench is built around the objects and projects you choose.

02 / Why it mattersClear tools. Better understanding.

A place to learn.
Room to do serious work.

The ambition is simple: make complex work approachable, while keeping the detail and control that experienced users need.

Curiosity belongs here

Start where
you are.

You might be exploring your first model or working through a professional dataset. Clear steps and visible results help you understand the work, rather than simply move through it.

Your judgment matters

Keep people
in the loop.

A prediction is a suggestion. Human-reviewed annotations are the foundation. BadBob makes space to inspect, correct and compare before deciding what comes next.

A connected workflow

Build on
what you learn.

Capture, annotate, curate, train and evaluate. The goal is a useful loop: each result helps shape better data and a more informed next model version.

The direction from here

BadBob is an evolving product. Continued development is focused on clearer workflows, dependable tools and the practical needs that emerge from real projects. Feedback and thoughtful collaboration are part of that direction.

See the workbench in action
03 / Experience behind the productEngineering × design

Johan Silvergrund

Technical depth.
A designer’s eye.

BadBob was created by Johan Silvergrund, a software engineer and UX/UI design engineer with more than 25 years of professional experience.

His work spans frontend, backend, APIs, databases and cloud infrastructure, including a decade working on business-critical enterprise software. Architecture, legacy modernisation and reusable platforms sit alongside a long-standing focus on the people using them.

BadBob brings those interests together: thoughtful product design, native development and a practical curiosity about machine learning.

PDF / English / 2 pages

04 / Open to what’s next

Good things start
with a conversation.

Have a project, a product challenge or an idea for BadBob? Johan is open to new projects and teams to work with, as well as conversations about partnerships and investment in BadBob’s future.

Tell us a little about what you have in mind.

Let’s talk.

A question about BadBob, a project or a possible collaboration.

20–5,000 characters. Please leave out passwords and other sensitive details.

Checking the spam protection…

Your name, email and message are used to respond to your enquiry. Cloudflare Turnstile checks for bots; Resend delivers the message. No newsletter sign-up.