Presentation: Building a Voice Assistant for Enterprise

Track: Modern CS in the Real World

Location: Pacific LMNO

Duration: 4:10pm - 5:00pm

Day of week:

Slides: Download Slides

Level: Intermediate

Persona: Architect, Developer, Front-end Developer, General Software, Technical Engineering Manager

This presentation is now available to view on InfoQ.com

Watch video with transcript

Abstract

Einstein Assistant is an AI Voice assistant for enterprises that enables users to "Talk to Salesforce". Users can dictate memos, update Salesforce records and create tasks using natural language. Einstein Assistant pioneers the use of Voice and Natural Language Processing (NLP) to enhance the user experience by reducing manual entry and increasing the timeliness and volume of data capture.

In this talk, we will go through the high-level architecture and workflow starting from Automatic Speech Recognition (ASR) on device to using NLP for identifying entities and intents in a single dialog conversation text.

Come to learn our practical approach to implementing a Voice Assistant and the unique challenges involved in integrating with enterprise data. We will discuss further opportunities to improve on our approach. For example, we look at adopting a more general multi-task learning NLP model (see decaNLP.com) instead of a single task model to enhance NLP performance.

 

Speaker: Manju Vijayakumar

Lead Software Engineer @salesforce

Manju is a full stack engineer at Salesforce, where she works on Einstein Voice Assistant, bringing the best in class NLP capabilities from Salesforce AI Research to production. Prior to Salesforce, she worked in a breadth of roles and domains and solved complex problems at Google, Goldman Sachs, NVIDIA and Akamai. She holds a Masters in Computer Science from Texas A&M University. She loves outdoor adventures, reading and blogging in her free time. 

Find Manju Vijayakumar at

Similar Talks

Scaling Patterns for Netflix's Edge

Qcon

Playback Edge Engineering @Netflix

Justin Ryan