grace - a deployment tool for deep learning models

Amith Reddy (~velutha)


165

Votes

Description:

A data scientist's job is usually to train a model often in the form of a jupyter notebook. However, to take this model to production takes different skills, a significant engineering effort and a lot of hidden technical debt accumulated over time.

Grace, a platform agnostic deployment framework addresses this problem (thus reducing the machine learning engineering effort) by acting as an orchestration tool to deploy deep learning models in production environment leveraging tensorflow serving , docker and kubernetes.

Any deep learning model to be deployed is configurable through a json spec containing input, output, model weights etc,.

Other services essential to maintenance like deep-dive monitoring tools, load testing tools, structured centralized logging are provided out of the box.

Prerequisites:

python, basics of machine learning/ deep learning

Speaker Info:

Venkat Karun is a full stack generalist and polyglot with 15 years of experience building high performance, distributed systems including a decade at Google. He enjoys reading up on functional programming and lambda calculus and tinkering with ev3dev and the lego Python ecosystem in his spare time. He is currently working as Chief Architect at NicheAI pvt ltd.

Venkatesh Mondi, an aerospace engineer by education worked in ISRO before finding his love for programming and machine learning. He worked as a software programmer in various platforms before co-founding NicheAI pvt ltd. He has been working on a variety of production grade computer vision solutions since it's inception. He can be found experimenting with gadgets, software, mathematics in his free time.

Speaker Links:

Venkat Karun

Venkatesh Mondi

Section: Developer tools and Automation
Type: Talks
Target Audience: Intermediate
Last Updated: