Walid Magdy

Image

What is your research focus?  

My research focus is in AI and Data Science, and more particularly in Natural Language Processing (NLP) and Computational Social Science (CSS).

I have around 20 years of experience in these fields working between academia and industry, where I started my research directly after graduation working for IBM research, then Microsoft research, where I learned to do applied research that leads to products currently used by millions of people. Recently, beside my work at UoE, I consult for industries such as the UN and Huawei to apply state-of-the-art tech in real-world applications.

During my journey, I have seen how AI and NLP have evolved from simple classifiers to the current era of large language models (LLMs), where my work has been always focusing in providing technologies that supports low-resources languages, such as Arabic my main focus, with similar and sometime better performance compared to English.

What is your innovation idea? 

Thankfully, it is much easier now to introduce my product than before, thanks to OpenAI that introduced chatGPT to the world while demonstrating its capabilities.

My product is not that far, but to summarise, imagine that I introduce another version of chatGPT with the following features:

1. Focus on low-resources languages, such as Arabic, while covering all dialects of the languages with the highest accuracies
2. Supports voice chat, not only text, and understands all accents and robust to low quality signals.
3. Provides two modes of operation:
   1. Closed-domain customer support, with Zero hallucination in response.
   2. Open-domain, with guaranteed minimal hallucination in response generation
4. Culture-aware LLMs, where our models take into consideration the cultural differences in the language compared to western culture.
5. Local-cloud and On-Prem hosting, where our models can be hosted locally in client’s cloud to protect privacy.

Why does this idea matter, what impact will it have on the world and what problem will it solve? 

One of the main usages of our technologies is in the domain of customer-support, where large companies who have large number of customers who require support can rely on our AI agents instead of real agents. This can be of large benefits for both the companies and their clients as follows:

• For companies:
  • Reduces the large amount of costs spent on customer support through call-centres or online-chats operated by humans.
  • Improve their KPIs through providing better and more efficient support to their customers.
• For Customers of the companies:
  • Less waiting time one phones or online chat to get an agent.
  • No need to go through a long tree of IVR to reach a choice required, but directly connecting with a human-like AI agent.
  • Dealing with AI agent that feels like human and can resolve up to 80% of the customer requests (95% of the most frequent requests).
  • Easy interaction through voice of text in their own dialects/accent.

Personally, and probably everyone, we suffer from the long wait to get support with many service providers we use, such as Telecom companies, banks, governments, airlines, e-commerce companies … etc. The experience is bad when introduced to IVR system that keeps taking us through many choices will reach our required request, then waiting tell getting in touch with an agent. Experience becomes worse when introduced to any AI supported service, which works badly in most cases. This becomes more severe for other languages or accents. My product resolves all of these issues.

What motivated you to apply to the Bayes Innovation Fellows programme and what do you hope to gain from it?

I am looking to explore routes to investment and better understanding of how to access the market that I am looking to enter. 

Links

The School of Informatics

The Alan Turing Institute