In the first episode of Behind the Interface, Jatin Leuva of Tcules and Samy Melaine of Eden AI look at UI/UX’s impact on AI product development.

There is no denying it: it is the golden age of Artificial Intelligence (AI). Industries across the board are counting on AI to drive unprecedented productivity gains, so much so that PwC expects AI to add $15.7 trillion to the global economy by 2030.

Given that the number of powerful AI models has exploded in the past decade, the real challenge to building these products is choosing the right ones for particular use cases, building the optimal workflows, and keeping it cost-effective. 

That is exactly where Eden AI comes in. 

In this Behind the Interface episode, Jatin Leuva, founder and CEO of Tcules, a design-led product development agency, talks to Samy Melaine, Co-founder and CTPO of Eden AI. We uncover the role of product development and user experience in building the next generation of AI-powered products, challenges and solutions to AI model selection, building workflows, and monitoring performance, Samy’s take on the best AI models for different use cases, and more.

But, before we uncover all that, we want to tell you about how it all started!

Samy’s big insight

Samy’s core observation, given his years of experience working with AI, was, 

AI was indeed going more and more from the hands of data scientists and AI scientists and into the hands of developers. We wanted to bet on that.

Now, what does this shift really mean?

According to Samy, you don’t need to be a data scientist (or a researcher, if we may add) to be able to build AI products or, for that matter,  use AI to its fullest potential.

That’s when Samy co-founded Eden AI: A full-stack AI platform for developers to efficiently create, test, and deploy AI. The platform has enabled 500+ companies to build AI products for various use cases, including marketing, sales, human resources, and customer support. 

Eden AI product philosophy

With so many AI models in the market, selecting the right one has been one of the biggest challenges for the users of AI products. As these models become increasingly specialized for specific functions, working with and integrating several of these into a single workflow also becomes a challenge.

This is why Samy and Eden AI are on a mission to simplify all aspects of application development, starting with model selection. “People across the world are on the lookout for the best AI models for some task… (On Eden AI) They understand they do not need to make a choice on just one model or provider. They can just have everything in one place.”

To improve accessibility to developers, Eden AI is essentially an API-first platform that “Anyone familiar with the concept of API can use…You don't need any AI expertise.”

Eden AI is, in fact, taking this even further. Instead of choosing their own AI models, users have been increasingly seeking recommendations for their use cases.

“We are building our recommendation system, that, for every input given, we will select, on the fly, the best model to handle it,” he said, adding that they can even prioritize recommending the most cost-effective ones (if that’s a criteria) to get the job done.

This single-minded focus on lowering the barrier to entry for AI usage is paying off in a big way, as Samy mentions that many of his thousands of users who were testing the waters on a single-use case are now expanding to build for many others with the platform.

‍

Marrying tech and product: Bringing out the best of both worlds

‍
When asked about Eden AI’s product development strategy, Samy spoke about how, in the initial days, ensuring the stability of the platform and bringing robustness was their top priority.

Given that Eden AI is a tool for developers, this puts the team’s developers in a unique position to understand their users’ needs well.  This is why when creating a new product or feature, the first step is to check the feasibility with the developers and then move on to refining the UX aspect. That said, user feedback has been equally important. 

“And we do emphasize on the fact that developers have to do some support themselves, so they are regularly interacting with the users,” he said.

As their tech stack evolved, Samy and the team began to focus on product development.

“We understood how complicated it might be for developers to understand some parts of AI since it's a new thing. We realized we had to simplify it and find an easy way to explain the value it can bring to users.”

He elaborates on this with an example of how Eden AI revamped the model selection process from mathematical evaluations through visual representation. 

“When we are orchestrating different models, we always give you visual inputs about the output at the end. So whether it's some transcription of the module, you can see the differences between the transcription visually or if it's an image or video analysis task, we can actually go in through your video and see the different outputs.”

This approach significantly improved the UX, but also intuitively gives a snapshot of the tone and the quality type of the final output, which is very important as Jatin from Tcules notes, a lot of these AI models operate in a black box model, where there is not a lot of transparency on how these models function.

This is where the consistent observability of an AI model’s performance also plays a big role in fostering trust among Eden AI’s developers. “Seeing how the model is behaving in production...for example, if it’s a GenAI model that generates text and if it's biased in its output, our objective is making it as clear as possible from week to week, whether the model’s behaviour has changed or not.” By constantly communicating this information, it becomes a powerful tool in enabling developers to make informed decisions while building their product on the platform.

Given that cost-effectiveness also plays a big role in model selection, the platform also prioritises showcasing the lowest-cost models that get the job done. 

‍

The role of UX in product development

Along their journey, Samy and the team have placed a lot of importance on user experience, and it was, in fact, Eden AI’s developers who advocated for it.

‍“They (Eden AI’s developers) were saying, ‘Hey, we actually love a robust product, but when we see a beautiful product, it makes it even better.’” For Samy, the biggest insight was that an improved user experience, whether it is about the ease of building workflows, the simplicity with which they can compare different models or just improved aesthetics, all contributed to increased trust for the user as it is a clear indicator how the people behind the platform pay attention to the details.

‍The most prominent example is how Eden AI worked with Tcules to reimagine their workflows. Earlier, their interface was fragmented, with as many as three different ways to create a workflow: templates, AI assistance, or starting from scratch.

With separate interfaces for building, testing, and configuration as well, it added to the complexity and cognitive load for Eden AI’s users, hindering their ability to use the platform to its fullest.

Rather than maintaining different interfaces for different actions, Tcules helped create a single, intelligent entry point that adapts to user needs. The interface was revamped to bring more user-friendliness to the workflow creation process by:

• Providing a single, clear starting point to process. 
• Enabling natural language search for templates.
• Offering AI assistance through mode switching.
• Maintaining all creation options while simplifying the interface.

‍Read also: Unifying workflow creation through intelligent search

Tcules’ Jatin also noted that there are some use cases within AI product development where UX professionals can also be natural allies without the need for any other intermediaries. He recounted how in Tcules AI lab, they got a few Machine Learning (ML) engineers to collaborate with UX designers as an experiment. The logic being that the team knew the final output needed for the user, and given that the UX person is naturally placed to know their users’ expectations, the overall model-building process would be more efficient.

“But the other thing that we did experience, which was very interesting, is the UX person was able to collaborate at the technical level where they were also writing better prompts with the ML engineer, and it somehow reduced the number of iterations that were going into the R&D,” said Jatin.

Samy’s tips

‍Eden AI was not the only subject of this conversation. Samy believes that after the Gen AI wave, reasoning in AI models is the next frontier in landscape. “A lot of people are talking about automated agents now… and if we want to have fully automated agents, we still need reasoning to be fully handled.”

For those interested in building AI-based applications, Samy’s advice is straightforward: understand what your clients want. “Don't do AI for AI’s sake, but do it for your user’s sake. Try to automate as many things as possible because that's the main job of AI: it's not to replace people, it's just to automate tasks for you and for your customers to gain more time and value as fast as possible.”

Till the time that reasoning in AI catches up, he also advises developers not to fully automate their work with one agent in the expectation it will handle everything itself.
‍
“Leverage different AI tools, build the logic yourself, understand what you want to do, and adopt AI in the right places.” He also had some opinions on the best AI models for popular use cases.  

I think Google is doing well with their generative AI models like Gemini. I also like Anthropic, as a lot of developers feel it's way better at code editing and understanding. I need to give a lot of credit to Amazon for their video analysis. For speech, ElevenLabs is doing a great job with speech-to-text or speech synthesis use cases...

The conversation ended on a high note with Samy’s advice for UX Designers. “Do not hesitate to ask GenAI models for ideas,” said Samy. “I've seen a lot of designers blocked with some task or the other, and they check out dribble for answers and try to get inspired from that. I really think that's something that should be done more frequently.”