Tell us a little bit about yourself and your journey thus far.
I graduated from the University of Southampton with a Master of Engineering degree in Computer Science. While at university, I had the opportunity to collaborate with computer scientist Sir Tim Berners-Lee on Semantic Web UX research as part of the Web Science Research Initiative. It was during this time that I met one of my fellow co-founders and Tray.io CEO, Rich Waldron, while he was a student at Bournemouth University.
Together, Rich, Dominic Lewis (Tray.io’s CRO) and I co-founded Tray.io with the mission to transform the fragmented business processes typically found in organizations into powerful outcomes through the power of automation, unlocking their full potential to solve challenges without the constraints of technology.
Today, Tray.io is a leading low-code integration and automation platform that bridges the gap between line-of-business workers and the complexities of code. We offer a unique solution that both non-technical and technical users can leverage to create sophisticated workflow automations that streamline data movement and actions across multiple applications.
Can you discuss the technical details behind Tray Merlin AI’s natural language automation capability? How does it leverage OpenAI technology?
Tray Merlin AI leverages OpenAI and works seamlessly with Tray.io’s connector, workflow and API technologies to automatically translate natural language inputs—prompts or requests written in plain English—into sophisticated workflows. Merlin uses a mix of GPT models—including 3.5, 4, Whisper and more—to process different parts of the natural language automation flow. Each of the models provide varying levels of capabilities, speed and fine-tuning, which are selected by Merlin to ensure the best user experience.
Think of Merlin AI as giving the LLM ‘brain’ a Tray ‘body’ that takes action on queries and builds the integration required to complete the business process—without passing customer data back to the LLM or requiring LLM training. While LLMs possess the capability to respond to any question almost instantly, nothing actually happens after the LLM responds. The burden is immediately on the person who asked the question, and it is their responsibility to take action on the response to achieve the desired outcome — and most times the action requires a high level of technical expertise. Now, Merlin can take that response and actually carry out the action on the user’s behalf.
How does Tray Merlin AI address the potential obsolescence of certain tech sectors due to the advent of LLMs?
Embracing digital transformation — which has now become mandatory in the ‘real time’ cloud-based reality — has come with the consequence of application and data overload. Organizations struggle with siloed information and a multitude of niche SaaS apps across every department. The demand for automation and integration has never been more critical. Until now, the antidote has been a modern, elastic iPaaS vendor, which would enable teams to ‘glue’ their systems together so their data can run in tune with the organization. However, this is no longer sufficient.
We’re facing a pivotal moment in the automation landscape, because the world has moved beyond the iPaaS architectures of the early cloud era. We’re recognizing that the requirements and expectations of iPaaS have shifted completely due to the advancements in AI. As a result, it’s important to deliver AI capabilities on a platform that is built for governance, security and scale. Those vendors who don’t yet have that will be challenged to get to the other side. In the case of Tray, Merlin AI works seamlessly with our connector, workflow and API technologies — and other platform capabilities including data transformation, robust authentication mechanisms and support for advanced business logic — to deliver a flexible, low-code and AI-augmented automation builder.
With this foundation, Merlin AI can be used by anyone — business technologists, front-line employees and developers alike — to develop fully baked workflows to execute their day-to-day tasks or retrieve information for specific business questions. In short, it removes the learning curve and lowers the technology barriers for users to build automated workflows. For the first time, business issues can be solved faster and more accurately and by a wider variety of people within the business through a natural language interface. The outcomes that can be achieved, and the fate of certain tech sectors, will be determined by what these modern-day iPaaS users will imagine to build with assistance from AI.
Can you describe the process of building out automation details with Tray Merlin AI’s low-code visual builder?
The process starts by simply asking Merlin AI — in the same way you would ask a colleague — to fetch an answer to a question or to develop an automated workflow. A user can type their request and Merlin will automatically build a workflow with the relevant business logic. For example, a marketing leader who is looking to add a new data enrichment source to a lead lifecycle management process can ask Merlin to create the necessary integration. Merlin will select the proper connector from the extensive Tray connector library, prompt for the required authentications (such as an enrichment product like Clearbit), execute the query and ensure that the results are appropriately reflected in the process moving forward. When finished, all the steps are displayed in the low-code visual builder through which the user can review and make any modifications if required.
How does Tray Merlin AI’s self-contained operational capabilities differ from other applications that interface with LLMs?
Merlin AI is the first iPaaS platform on the market to offer native generative AI capabilities that users can leverage to securely automate complex business processes. Unlike other applications that interface with LLMs, the operational capabilities of Merlin and the underlying Tray platform are self-contained, meaning Merlin only needs to fetch small pieces of information from the LLM on an as-needed basis during the integration building process. As a result, customer data is never exposed or sent to the LLM.
Additionally, Tray Merlin AI works across the entirety of the customer’s software stack, which is very different to most of the GPT-related chatbot announcements we’re seeing that, at best, will only be able to take pre-defined actions within their own applications.
How does Tray.io’s connector, workflow, and API technologies integrate with Tray Merlin AI’s generative AI capabilities?
There are three fundamental traits of the platform that makes Tray uniquely designed for AI-powered automation:
- First, the Tray Merlin AI intelligence layer is built into our foundational core of connectors, workflows, APIs and low-code interface. Merlin automatically detects the necessary connectors needed for an integration and seeks authentication when required. All the steps that Merlin builds are displayed in Tray’s low-code visual builder, which allows the user to easily review and make any modifications, if necessary.
- Second, our 600+ “out-of-the-box” connectors use a standardized input/output schema that makes it simple to integrate with AI systems. The input schema supports validation enabling us to fine-tune inputs for connectors based on provided validation messages. Validating inputs before sending requests to third-party services enhances the quality and consistency of integrations, ensuring no incomplete requests are made. This means the LLM doesn’t need to manage inconsistent error responses from third-party APIs.
- Finally, our platform lets users create workflows that are exactly like their business processes. With features such as unlimited testing, branching, loops, callable flows and triggers, Tray lets anyone harness the power of AI and automation to build powerful workflows with advanced business logic.
What are some of the technical challenges you faced in the development of Tray Merlin AI, and how did you overcome them?
This is actually the work I find most interesting, the opportunity to work with exciting new technologies, unpacking the nuances and figuring out how to create the best experience for our customers. As we began the work on Merlin, there were been plenty of technical hurdles to overcome, but the most impactful was really how to curate and harness the end-user experience. LLM’s present so many possibilities, and a lot of our work has gone into ensuring we are providing the right guardrails for our customers so they get the productivity gains that are available to them.
In your view, how will Tray.io continue to innovate in the low-code automation space in the upcoming decade?
We founded Tray.io with the mission of empowering and arming all employees with the tools they need to execute day-to-day tasks through enterprise-wide automation, and we remain committed to this mission. The release of Merlin AI accelerates our vision to lower the barriers that prohibit enterprise-wide automation and truly empower all employees. Going forward, we will be strengthening the versatility of our platform so that our users can mix and match the various solutions Tray offers in a seamless manner. Cross-pollinating capabilities across Tray’s portfolio equips everyone in the enterprise to leverage what’s best for their needs and achieve accelerated automation outcomes. For instance, if automation has already been built using our low-code interface, a developer could leverage that capability and embed it in code instead of recreating that functionality from scratch using APIs.
Tell us about your favorite book and why did it stay with you?
My favourite book probably bucks the trend of ‘nonfiction’ business books. I’m a big Michael Crichton fan and find his work blends the much needed distraction of fiction with a lot of science. I particularly enjoyed ‘Timeline’, as it gave me an insight into quantum theory that I would have struggled to gain from reading a textbook.
Any quick advice for technology enthusiasts?
My advice is to stay curious. There are so many amazing technologies out there just waiting to be tapped into. Take the time and have fun experimenting with AI—ask questions, push boundaries and find out what’s possible today and in the future in terms of how you can use AI to deliver better outcomes.