BT taps AWS for internal GenAI platform

BT's Digital Unit has launched what it calls the GenAI Gateway, a platform giving it access to a wide variety of large language models (LLMs) from the likes of Anthropic, Meta, Cohere and Amazon, naturally.

BT can leverage these for any use case that springs to mind, but it must be noted that AWS is doing a lot of the heavy lifting here.

The platform has been developed using Amazon Bedrock, which is a suite of GenAI application development services, and Amazon Sagemaker, a managed machine learning (ML) service that enables enterprises to train and customise ML models for any number of use cases. AWS is also providing professional services to BT to get it all up and running.

GenAI Gateway's first beta use cases are already up and running. BT's networks arm Openreach is using it to summarise engineering notes on Ethernet and full fibre jobs, helping to simplify processes and boost productivity for its teams as well as CSP customers.

BT's business, legal and procurement teams have also begun using the GenAI Gateway to improve contract analysis.

Rather than test the waters or dabble with a range of different AI solutions, GenAI Gateway appears to be an 'all-in' play by BT.

It made very clear in the announcement that deploying one platform will reduce the risk of duplication of effort and resources. It also means that APIs, security configuration and infrastructure management can all be managed centrally, mitigating the risk of error or maintaining separate LLMs for different use cases.

On a related point, BT also notes that ad-hoc use of multiple LLMs is good for test and development, but not for large-scale use. Adopting a flexible solution that supports multiple models means BT's teams can use the right model for the right use case without driving up costs and inefficiencies.

In addition, a multi-LLM approach provides safeguards in the event that one of them starts making up false information (known as hallucinations) or stop behaving as expected.

GenAI Gateway also applies several guardrails, enabling it to tick that all-important 'ethical AI' box.

These include filtering out personal identifiable information and using data located within the UK, as well as isolating trained models from one another to stop them sharing data too freely. BT's GenAI applications have also been hobbled somewhat in an effort to limit the risk of toxic interactions, and ignore queries that are unrelated to their specific use cases.

"AI is helping us reimagine the future of our company. We believe that where our data is a constant, we need flexibility with our LLMs. GenAI Gateway allows us to tap into this powerful new set of technologies at scale, in a way that is safe, responsible, flexible and scalable, delivering the ambition we have for AI to unlock the human potential within BT Group, today and in the future," said Deepika Adusumilli, managing director, data and AI, at BT's Digital Unit.

"The BT Group GenAI Gateway is showcasing how enterprises can effectively deploy generative AI at scale and speed," added Fabio Cerone, general manager EMEA telco at AWS. "It's been a brilliant, pioneering opportunity to collaborate and work backwards from the customer to provide a way to accelerate deployment of generative AI use cases into production with embedded security and compliance. The GenAI Gateway will trigger the flywheel effect in the adoption of generative AI, delivering quicker results for BT Group and its customers."

A large-scale deployment of GenAI by an incumbent operator is a noteworthy milestone in AI's journey towards mainstream adoption and credibility and so-on. However, the use of multiple LLMs is likely to put an upper limit on the size of the revenue opportunity for these individual AI companies, which could prove significant in a market that demands massive up-front investment in infrastructure.