Let's talk
Back to products

The work, in one line

Making computer-vision workflows legible — for users, teams, and the market.

For users

Clearer workflows across data, training, deployment, and monitoring.

For the team

Personas, workshops, and shared language for priorities and product direction.

For the market

Launch storytelling, demos, releases, and campaign assets that made the product easier to understand.

Business context

$30M Series A

Ultralytics announced a $30M Series A during a broader push to expand computer-vision R&D, enterprise solutions, and platform work.

My work sat across the product and storytelling layers of that platform expansion.

ElephantSquareOne
Public announcement ↗

01 Four connected layers

  1. 01

    Product understanding

    Personas, workflows, team structures, and opportunity framing.

  2. 02

    Product experience

    Interfaces for data, training, deployment, monitoring, and live model behavior.

  3. 03

    Team alignment

    Workshops and facilitation to clarify priorities and turn ambiguity into direction.

  4. 04

    Product storytelling

    Launch videos, releases, decks, demos, and campaign assets that helped explain the product.

02 The product was a workflow system

A learning loop, not a linear flow

Not a single interface. Users move through a loop — prepare data, train models, deploy and monitor them, then feed real-world signal back into the data. My work helped make that loop easier to understand, design, and discuss.

03 From ambiguity to direction

A persona & workflow framework

The goal was not to document personas. It was to give the team a shared language for product priorities — workflow archetypes, from individuals to enterprise teams, that the product had to serve.

IndividualStudent / ResearcherLearning computer vision and experimenting with models.
IndividualIndependent specialistAdapting and deploying models for specific client problems.
StartupStartup ML implementerBuilding computer-vision features into a young product.
TeamProduct engineering teamShipping and maintaining CV capabilities across a codebase.
EnterpriseEnterprise AI / CV teamRunning vision models at scale, under review and governance.
OperationsData annotation specialistPreparing and labelling datasets with accuracy and consistency.

04 Designing the core workflows

Data

Helping users prepare better datasets.

Collect, upload, annotate, and review data — with AI-assisted and manual workflows designed as part of one quality loop.

Training

Making model training observable and actionable.

Users need to see what is running, compare performance, and stop experiments before they waste time and compute.

Deploy & monitor

Closing the loop between real-world behavior and the next decision.

Once a model is live, teams need to read performance, detect issues, and turn those signals back into better data and better models.

The through-line

The design challenge was to make each stage visible enough for the next decision.

05 Post-launch workflow audit

Closing the feedback loop

After launch I audited the live workflows end to end — where users stalled, where the loop leaked — and turned the findings back into data, training, and monitoring decisions.

End-to-end workflow audit board — every screen, stall point and follow-up note in one map.

06 Explaining the product outside the interface

A complex AI product has to be understood in more places than the UI.

Launches, demos, release notes, growth campaigns, and company milestones all had to explain the same product logic clearly.

Launch hero
Feature release
Feature highlight
Company milestone
Vision & messaging
Impact in numbers
Brand mark

07 What changed

Ultralytics

Connected product areas

Data, training, deployment, and monitoring became easier to discuss as one workflow loop.

Linear

Clearer backlog and priorities

Workshop outputs and workflow audit findings were translated into structured follow-up work in Linear.

Social

Better product communication

Launch, release, and growth materials explained complex AI capabilities in a more consistent product language.

Workshops

Stronger decision-making

Personas, workflow maps, and audit artifacts helped the team align around where to focus next.

08 Reflection

Product design in AI is often about making invisible systems discussable: model behavior, user maturity, team workflows, constraints, and the story around the product.

The most useful artifacts were not always the prettiest ones. They were the ones that helped the team make a better decision.

Working on a product with this kind of complexity?