How AWS Inferentia Enables Autodesk to Expertly Answer Over 100,000 Questions Per Day | WIRED Brand Lab

[Narrator] Nowadays, virtual assistance or chatbots

play an increasingly important role

in providing quality customer support.

But what happens when a company like Autodesk,

needs their chatbot, Ava, to serve a variety of industries,

each with complex technical language as fast as possible?

[gentle music]

When asked a question,

Ava uses a natural language processing based

machine learning model

that filters words through a set of parameters

to output an answer.

But serving different industries at once

comes with serious cost, speed, and accuracy challenges.

That's why Ava is now powered by AWS's Inferentia.

A custom silicon chip,

purpose built for machine learning inference workloads.

This helps Ava route questions

through multiple ML models at once

and generates a response in milliseconds.

Ava can now serve five times as many customers

with increased accuracy at almost half the cost

of GPU powered compute instances.

With this high performance, low cost, and ease of use,

the future for chatbots efficacy is limitless.