Managing mobile AI projects at Hive AI (2018 - 2021)

As company transitioned to focus fully on enterprise AI, I managed development of different types of machine learning applications on mobile and became the sole manager for the engineering team in New Delhi.

Selected sub-projects

Hive Work, Mobile Product Owner (2018 - 2020)

Data labelling jobs where workers are paid per task, and the data is used to train machine learning models.

Screenshots

1242x2688bb.png{:width 240} 1242x2688bb-2.png

1242x2688bb-3.png{:width 240} 1242x2688bb-4.png

Hive AI Model Demos, Product Owner (2018 - 2022)

Fast-paced model prototyping to get MVP to business team and into client hands to show capabilities.

Visuals

Released app screenshots

Receipt parsing

receipt video from march 23, 2020?

Logo and object detection

Using TensorFlow Lite
logo-detection-demo-tflite.png

Depth detection

Wireframes for testing NSFW/Violence model on device (drawn with Balsamiq)

Use screen recording video as source to test model
hive-demo-wireframe-v1.jpg

Use user selected images as source to test model
hive-demo-wireframe-v1-2.jpg

View test results
hive-demo-wireframe-v1-3.jpg

Scribe, Product Owner (2020)

Using a custom built Hive speech-to-text model to build a transcription app to record audio and transcriptions side by side.

Converted existing Hive Work a desktop web product to a mobile experience from product design to release.

Designed the experience of drawing a bounding box on a small screen for Hive Work. The user needed to have very fine tuned control, but the steps needed to be streamlined so the user could do tasks quickly.

Bounding box functionality demo:

Took Scribe from being a proof of concept for the speech-to-text model to expand functionality and provide utility for consumers.

Built wireframes and working demos to demonstrate Hive’s models and APIs.

Designed the software architecture for shared libraries to reuse our on-device model framework.

Worked with infrastructure team for the data labeling pipeline that required large amounts of data processing.

Managed a fully remote team in a different time zone: 4 engineers, 2 designers, 2 testers. This also required changing internal processes to better support the new team environment.

The wellbeing of the New Delhi office was often neglected by the San Francisco team, so I worked to create a healthier and more inviting environment for the New Delhi team. This meant both being a manager that listened with warmth and sympathy, as well as making sure that the team was kept in the loop about executive level discussions and priorities.

Grew the Hive Work mobile apps to over 600K total users with 5M tasks completed daily. Mobile platform went from 0% to 20% of Hive’s data labelling workforce over 4 months.

Coalesced priorities from executive, business, machine learning, and mobile engineering teams to build quick high value prototypes for many different in-house models. This also meant often having to quickly build proof-of-concept demos myself if the CEO needed something quickly for a pitch.

Found opportunities to take advantage of new, yet to be pervasive technologies. TensorFlow Lite and PyTorch Mobile were still experimental forms of shrinking machine learning model sizes for use on mobile devices.