User:Lauragrainger/sandbox/PolyAI

PolyAI is a UK-based company developing enterprise voice assistants for customer service using conversational artificial intelligence. The voice assistants use proprietary machine learning and natural language processing technology that has been trained on over a billion conversations, allowing it to scale across industries, use cases and world languages.

History
PolyAI was founded in 2017 by Nikola Mrkšić, Pei-Hao Su and Tsung-Hsien Wen. The three founders met at the Machine Intelligence Lab at the University of Cambridge where they studied under Steve Young. Before founding PolyAI, Mrkšić worked at VocalIQ, a Cambridge speech technology startup that was bought out by Apple Inc. in 2015 for $100m and became an essential part of Siri.

In 2019 the company raised €10.7 million in Series A funding in an investment round led by Point72 Ventures, with participation from Sands Capital Ventures, Amadeus Capital Partners, Passion Capital, and Entrepreneur First.

The company was awarded Company of the Year 2019 by the Cambridge Computer Lab Ring.

In November 2020, PolyAI was named a Gartner 2020 Cool Vendor in Conversational AI Platforms.

PolyAI was named to the 2021 CB Insights AI 100 List of Most Innovative Artificial Intelligence Startups, an annual list of the 100 most promising private AI companies in the world.

ConveRT
PolyAI developed a proprietary dual-encoder model, ConveRT, released on the public domain in April 2020. The model uses subword representations and light-weight transformer-style blocks to encode text, resulting in a model that can be quantized to under 60MB. In October 2020 Chien-Sheng Wu and Caiming Xiong from Salesforce published results from their research on language models, evaluating their ability to encapsulate conversational knowledge in application to Conversational AI tasks. Results showed that PolyAI’s ConveRT model outperformed all other models in three out of four of the classification probe evaluations. The ConveRT model was taken down from the public domain in September 2020.

ConVEx
In October 2020 PolyAI published a paper on their conversational value extractor, ConVEx. ConVEx is an efficient pre-training and fine tuning neural approach for slot-labeling dialog tasks, a critical natural language understanding component of task-oriented dialog systems.