Product Design & Development

A time-series data collection & annotation platform to help build edge ai models.

Bring ambitious, sensor-driven machine-learning products into production through integrated data collection and testing. This was an internal product we developed while providing model development and data collection services for Ford.

My Role

I put together a team of four consisting of myself, a backend engineer, frontend engineer, and embedded engineer. I led the product design, development, and implementation of the entire tool.

Challenge

Today, AI too often is built in isolation. Teams build models, deploy them to the edge but have little insight into how they perform in the real world and if they work for their users in every edge-case. Our machine learning and data collection teams struggled to manage large amounts of training data from field tests and recordings. We wanted to build a tool that provided everyone with the same source of truth.

Solution

A dashboard and companion app that allowed both teams to take a glance at the quantities and status of the entire dataset. As well as sample playback tools and annotation interfaces that allowed for our QA team to crop and analize time series data.

Outcomes

Playground Studio was used by our team in collaboration with Ford to manage, operationalize and clean over 100K samples of machine learning training data.

Playground Studio connects seamlessly with development hardware devices to collect audio, motion, and image data for machine learning model development. Putting all the data in one collaborative interface to remove the gap between data collection experts and machine learning engineers.

Playground Studio Overview

A demo of the entire data collection and annotation process. Studio can easily connect with development kits such as RaspberryPi to pull and store annotated recordings. Meta data can be configured and adjusted based on the data being collected. Multiple sensors (audio, motion, image) can be recorded at one time to make it easier to apply a single annotation across all.

A demo of the entire data collection and annotation process. Studio can easily connect with development kits such as RaspberryPi to pull and store annotated recordings. Meta data can be configured and adjusted based on the data being collected. Multiple sensors (audio, motion, image) can be recorded at one time to make it easier to apply a single annotation across all.

Field Kits

We made a small batch of flexible field recording kits that could be used to collect audio, motion, and image data from any environment. These served as development kits prior to using production hardware.

We made a small batch of flexible field recording kits that could be used to collect audio, motion, and image data from any environment. These served as development kits prior to using production hardware.

The Team

We worked with our machine learning and data collection teams to understand their biggest gaps and opportunities for the tool and collaborated with our clients at Ford regarding their data collection and management issues.

We worked with our machine learning and data collection teams to understand their biggest gaps and opportunities for the tool and collaborated with our clients at Ford regarding their data collection and management issues.