layout copilot mockup
gradient background

layout copilot

I created a single screen design concept focused on utilizing state of the art machine learning technology, human-centered Human-in-the-Loop interaction, and my published model training framework.

/

impact

This concept design completely reimagined the way that users set up layouts in Hyperscience, turning a 100% manual process into a low-touch, machine learning assisted, fast and easy experience.

 

Layout Copilot is the foundational experience for Automation Strategy, starting with automation for all use-case setup steps.

/ role

Concept design

/ problem

Make creating layouts a breeze

The previous experience for creating layouts was entirely manual - users were expected to manually source template documents, manually define each data extraction point, manually define data types, manually define output transformations and manually define the entire output format.

 

Hyperscience customers found this manual process to be tedious and expensive.

 

"The product claims to use machine learning for automating processes, so why am I doing so many things manually?"

 

The Hyperscience product was using machine learning for very specific document processing and text-extraction tasks, but not applying that same technology across the end-to-end business process automation workflow.

/ solution

Agentic layout building

I applied my published model training methodology to the identified problem with layout building, turning each manual step into an automated, AI assisted, low-touch experience.

triage

bullet point

import disordered dataset

bullet point

model naïvely clusters data

bullet point

human corrects cluster assignments

organize

bullet point

human organizes the identified clusters into hierarchy

bullet point

model learns hierarchy within clusters

bullet point

human corrects hierarchy

automate

bullet point

model automatically identifies and categorizes new incoming data

bullet point

corrected samples retrain the model in an active learning loop

human supervision for low confidence samples

human quality assurance for model accuracy measurement

conditional HitL

layout copilot mockup

/ selected artifacts

/ feature highlight - keyboard shortcuts

An accessible, keyer persona streamlined experience

gradient background
megaldonatron icon

- ChatGPT

upon checking again, there are indeed two ‘r’ characters in the word ‘strawberry’