The Technology Interview: Rahul Pathak, VP Data & AI GTM AWS
The race to implement generative AI in enterprise environments has revealed a fundamental truth: access to powerful AI models is no longer a differentiator. What matters now is the ability to integrate those models with proprietary business information.
“Today, anyone with a credit card can access state of the art AI – the real differentiator is when you can combine that AI with the unique data you have about your business and customers,” says Rahul Pathak, Vice President of Data & AI GTM (go-to-market) at AWS.
In his role, Rahul works with organisations worldwide on their AI implementation strategies, with particular emphasis on the data foundations necessary for success. His background includes helping develop key AWS data services that now serve as infrastructure for AI applications.
“I collaborate with global specialists, sales, and partner teams to ensure our customers can effectively harness their data and AI capabilities, enabling them to create real value for their businesses,” he explains. “For instance, by using AWS’ purpose-built services, such as Amazon Redshift and Amazon Aurora – products I helped develop earlier in my time at AWS – customers can build robust data pipelines and databases that fuel generative AI solutions.”
This integration between established data infrastructure and newer AI capabilities is central to AWS's competitive positioning. “Data is at the core of every AI initiative, especially generative AI, where the quality and relevance of data directly impacts the accuracy, creativity and reliability of AI models,” he says.
Amazon Nova models transform creative workflows through multimodal capabilities
The company’s multimodal Nova foundation models, announced at the 2024 edition of AWS re:Invent in December, process text, images and video prompts simultaneously, with early adopters already implementing these capabilities in production environments. Rahul highlights digital marketing firm Dentsu as a prime example of creative workflow transformation.
“Dentsu is integrating Amazon Nova Reel – a state-of-the-art image generation model that creates professional-grade images from text or images provided in prompts – into its creative process,” he notes. “Amazon Nova Reel reduces the overall time it takes to generate new assets from weeks to days.”
Enterprise software provider SAP is another company integrating Nova models into its AI infrastructure. “SAP is integrating Amazon Nova models into its SAP AI Core generative AI hub’s family of supported large language models,” Rahul explains. “This enables developers to create new skills for Joule, SAP’s AI assistant, and securely build AI-driven solutions that harness the full business context captured in SAP data, enabling automation, personalisation and advanced solutions like supply chain planning.”
AWS Trainium chips reduce AI costs while improving performance
The escalating computational demands of advanced AI models have prompted AWS to develop its own accelerator hardware. As Rahul explains, this custom silicon development is a direct response to customer requirements for both performance and efficiency.
“As models grow in size, they are pushing the limits of compute and networking infrastructure as customers seek to reduce training times and inference latency,” he notes. “Even with the fastest accelerated instances available today, customers want more performance and scale to train these increasingly sophisticated models faster, at a lower cost.”
The Trainium chip family delivers meaningful efficiency gains over traditional GPU infrastructure, with Rahul citing specific performance metrics. “In practice, Trainium2 delivers 30-40% better price performance than traditional GPU-based servers, with the power to handle billion-parameter AI models.”
Energy efficiency represents another critical advantage, particularly as AI deployments scale. “Trainium chips are also very power-efficient which means they help significantly with sustainability. Trainium2 instances are designed to be three times more energy efficient than Trainium1 instances.”
Looking ahead, Rahul outlines further performance improvements in the roadmap. “Trainium3 chips will deliver four times the performance of their predecessors, allowing customers to build bigger models faster and deliver superior real-time performance when deploying them.”
Enterprise AI adoption requires focus despite implementation barriers
As organisations move beyond experimentation, Rahul identifies several interconnected barriers to production AI deployment. Strategic focus tops the list, with many organisations struggling to identify the most valuable use cases.
“One challenge businesses face is identifying where to focus their generative AI efforts for the most impact,” he says. “It’s important to cut through the hype, and focus on the business case for generative AI.”
Data infrastructure readiness forms the second critical foundation, with Rahul emphasising the importance of bringing AI to existing data rather than the reverse. “Another challenge is getting the data foundations right. The key is to take generative AI to the organisation's data, not taking its data to generative AI,” he explains. “Organisations that have a clear data strategy and are already invested in a data foundation in the cloud have a strong advantage when it comes to adopting AI.”
The technology skills gap represents the third major constraint on implementation efforts, with demand for AI talent far outstripping supply. “Organisations are struggling to find the right talent to harness the potential of Gen AI. Research from Access Partnership and Amazon Web Services shows that 73% of employers say hiring AI skilled talent is a priority, but three out of four who consider it a priority can't find the AI talent they need.”
Amazon Q Developer transforms software engineering productivity
Among the most immediately valuable applications of generative AI, according to Rahul, are AI-powered development assistants like Amazon Q Developer, which demonstrate substantial productivity improvements by automating routine coding tasks.
“We are seeing that AI-powered assistants for software development are making it easier and faster to build, secure, manage and optimise applications on AWS,” he explains. “The benefit of this is that AI does the heavy lifting such as the tedious tasks that take up time, and allows developers to do more with the tools they are using.
“For BT Group, over 100,000 lines of code have been generated by Amazon Q Developer in its first four months – automating around 12% of the tedious, repetitive and time-consuming work. The solution provides 15-20 suggestions of code per active user per day for BT Group, with an acceptance rate of 37% by its software engineers.”
Internal AWS metrics suggest even greater potential, Rahul reveals. “Early indications signal Amazon Q could help our customers and employees become more than 80% more productive at their jobs; and with the new features in our roadmap, we believe this will only continue to grow.”
AWS implements framework for responsible AI development
AWS has embedded responsible AI practices throughout its development process, with Rahul outlining a comprehensive approach spanning technical safeguards and employee education.
“Responsible AI is a longstanding commitment at Amazon. From the outset, we have prioritised responsible AI innovation by embedding safety, fairness, robustness, security, and privacy into our development processes and educating our employees,” he explains.
Recent product releases focus specifically on ensuring AI systems operate within appropriate boundaries, with Rahul highlighting tools released earlier this year. “In April 2024, we announced the general availability of Amazon Bedrock Guardrails and Model Evaluation in Amazon Bedrock to make it easier to introduce safeguards, prevent harmful content, and evaluate models against key safety and accuracy criteria.”
The company has subsequently added features to address hallucination issues in Gen AI systems. “We recently added contextual grounding checks in Guardrails to detect hallucinations in model responses for applications using RAG and summarisation applications,” Rahul explains. “Contextual grounding checks can detect and filter over 75% hallucinated responses for RAG and summarisation workloads.
“To date, we have developed over 70 internal and external offerings, tools, and mechanisms that support responsible AI, published or funded over 500 research papers, studies, and scientific blogs on responsible AI, and delivered tens of thousands of hours of responsible AI training to our Amazon employees.”
Enterprise AI applications expand from internal to customer-facing solutions
Enterprise AI implementation is shifting from internal productivity tools to customer-facing applications, with Rahul observing increasing confidence in deploying AI in production environments.
“Initially we saw customers starting out with internally facing employee apps – in part because it can be lower risk, in part to boost productivity coming out of Covid-19 workforce disruption,” he explains. “Now customers are starting to feel more confident with real-world, external customer-facing apps and services.”
One such customer, sports organisation PGA TOUR, is one example of how Amazon is powering new, interactive fan experiences. “The PGA TOUR is using Amazon Bedrock and its proprietary data to build an interactive experience for golf fans to find information of interest about PGA TOUR events, players, stats, or other type of content, including generated video clips highlighting TOUR footage that is relevant to the answer provided,” he says.
Beyond content summarisation, organisations are increasingly building AI agents capable of taking autonomous action, representing a significant evolution in application capability. “As organisations make Gen AI a core part of their applications, we are also seeing that they want to do more than just summarise content and power chat experiences. They also want their applications to take action,” Rahul explains.
“AI-powered agents can help customers’ applications accomplish these actions by using a model's reasoning capabilities to break down a task, like helping with an order return or analysing customer retention data, into a series of steps that the model can execute.”
This multi-agent approach enables complex tasks to be distributed across specialised components, with Rahul outlining potential financial sector applications. “Using multi-agent collaboration in Amazon Bedrock, customers can get more accurate results by creating and assigning specialised agents for specific steps of a project and accelerate tasks by orchestrating multiple agents working in parallel. For example, a financial institution could use Amazon Bedrock Agents to help carry out due diligence on a company before investing.”
To read the full article in the magazine, click HERE.
Explore the latest edition of Technology Magazine and be part of the conversation at our global conference series, Tech & AI LIVE.
Discover all our upcoming events and secure your tickets today.
Technology Magazine is a BizClik brand