x600 How DynamicNLP Creates Virtual Assistants in Minutes with Yellow’s Raghu Ravinutala

How DynamicNLP Creates Virtual Assistants in Minutes with Yellow.ai CEO Raghu Ravinutala

Conversational AI solutions are becoming must-have customer experience options for enterprises, but they often require gathering synthetic data and months of model training prior to launch. Yellow.ai processes over 2 billion conversations every quarter, and the company used that data to create what they call DynamicNLP. That’s dynamic, as in created in the moment, and NLP, as in natural language processing.

The solution is based on zero-shot learning techniques and enables companies to go live with a conversational chatbot or voice assistant in a few days with 97% accuracy instead of training and retraining language models over many months. In fact, Raghu Ravinutala, CEO of Yellow.ai, says that you can go live in minutes. The launch time in days reflects the typical process of integrating with backend systems that enable the virtual assistant to complete more complex tasks.

Zero-Shot Learning

The innovation behind DynamicNLP is the implementation of zero-shot learning. GPT-3 has become famous for its successful implementation of few-short learning. That means the model requires less training than traditional approaches. It requires a few “shots” at learning from previously labeled data.

Zero-shot learning removes the formal training dependency. It doesn’t require any synthetic data or formal training. Ravinutala says Yellow.ai was able to accomplish this because it already had access to billions of conversations. There is a good chance that your user’s question or something very similar has already been answered. Yellow’s conversational model is capable of identifying matching requests and similar requests. That means you don’t need to seed your implementation with data.

This would normally be characterized as a pre-trained model. However, Ravinutala suggests we don’t use that term. Pretrained models assume some manual training inputs. They also are static between training sessions. Yellow’s DynamicNLP is a self-trained model that continually updates as new conversations are processed. So, it’s not pre-trained in a traditional sense, nor is it static.

A Quick Rise in Accuracy

Yellow is making the new solution available to all of its existing users and new customers today. Existing users will immediately experience a more robust virtual assistant that produces fewer errors after adopting DynamicNLP.  Ravinutala says that a large airline recently updated its virtual assistant to DynamicNLP and saw a 20% increase in accuracy in just two days.

More About 10 Minutes On Interview Series

This interview is part of a new series called 10 Minutes On by Voicebot.ai. The interviews focus on a single topic, are short enough to watch between Zoom meetings, and long enough to share interesting insights with some depth and detail. You can find more video interviews like this on Voicebot’s YouTube channel or by clicking Videos in the website’s top navigation bar.

Enterprise Conversational AI Startup Yellow.ai Raises $78.15M

10 Minutes On Amazon’s Real Reason for Acquiring iRobot with Voicebot’s Bret Kinsella (Video)

Raghu Ravinutala CEO of Yellow AI – Voicebot Podcast Ep 237