OpenAI goes open source: Nearform experts share how enterprises can benefit - Part 2OpenAI goes open source: Nearform experts share how enterprises can benefit - Part 2OpenAI goes open source: Nearform experts share how enterprises can benefit - Part 2

Last week, two Nearform experts sat down to share their perspectives on the implications of OpenAI launching its models to the public, releasing gpt-oss as a free download.

Sarah Michelle Egan, Technical Director based in the US, and Colm Harte, Head of Delivery for Europe, based in Ireland, each bring years of hands-on experience in the open source community, as well as expertise in collaborating with ambitious enterprises to accelerate their AI adoption. Their discussion ranged across a number of topics related to open source and AI, including a look ahead into what decision-makers can now take into consideration as they make challenging decisions around their digital, and business, model.

SE: The performance metrics [of Open AI’s product] are not quite to the level of the proprietary models in most cases. With any open source tool or any vendor that you're analyzing for adoption into your software, or your toolkit, there are a lot of considerations, especially with AI models. Understanding security and compliance, and designing access from the beginning, asking yourself if the open source model can handle the expected production loads that you're expecting, and governance and monitoring again. I think companies will have to consider how they can build those controls for drift latency, and how the model evolves over time as they're using it.

And then also managing data is a huge part of it. With a lot of AI projects, companies are getting really excited about adopting AI and then realizing their data is not in the right place. And so we'll probably see a lot more effort being put into ensuring that their data is in a good spot to be used with these open source tools and then building the pipelines to appropriately funnel it where it needs to go, so it can be utilized.

CH: If you have, say, sensitive data, you probably have different people in the organization who can see different data, and you have all the needed controls in place. As part of feeding that data now into a large language, you also have to think about how you're still going to maintain the data restrictions that you need to have in place. Because you can't necessarily have everybody going to the LLM and all being able to now query all the data. You have to take all those considerations into account as well, and how do you plan and manage that from a data governance perspective.

I think that's you then have to look at your own capabilities and actually determine if you have the capability to stand this up properly in a production environment. Again, there's all these pieces now you have to manage yourself, so you have to plan. You've got all the DevOps considerations, and have to take on the security considerations yourself. And then how are you going to maintain it over time? How are you going to deal with it when there's an issue in production? You're going to need to support it properly - it takes a lot more effort to run this yourself; it's not as trivial as just hooking into an API set.

That's why you would bring in a partner like Nearform that has experience in this space, knows how to do these things, has the history of building production systems, knowing how you create scalable, secure, reliable production systems, and that this is just another part of a solution that you can bundle into that.

SE: We partner with our clients, end-to-end, to help them select the right model. We optimize infrastructure, deploy with secure pipelines, and then with any tool - and especially open source AI - we will look to tailor the solution to fit their business objectives and really be pragmatic about their implementation. So we keep cost and reliability in mind, but also long-term maintainability.

I think with open AI behind this, and when you see a lot of open source projects that have large enterprise companies behind them, it's less of a risk because there is more backing, for it to be maintained for a long time.

We know that we'll “hand off” outputs of a project to a client eventually, and we want to make sure that they're in a good state, ready to maintain and keep the code and run their systems efficiently for the long-run.

CH: We have clients that have concerns about using proprietary models. What's happening with their data? Is it going to get used for training? Or they might have a lot of sensitive data. They're just not comfortable sending it to another company for that sort of processing.

The advantage of this is you can now do everything within your own environment. You've got full control, full security. You can manage it better from a governance perspective, especially when we're dealing with sensitive data. You need that reassurance that there’s not going to be any breach of data, from a leakage perspective.

And then the fact that you're getting the level of capabilities that are in the newer models is exciting. Again, as an open source model that you can fine tune and train and tailor it to your needs. It just gives a lot more flexibility into the market, which is always good from either enterprise or consumer perspective, right? You just have more options. But it is a trade-off: yes, you can have more control and you have better management of your own data and who can access it, but you have to do a lot more heavy lifting yourself.

At Nearform, we've been building up a lot of expertise in this area. We have a long track record and history of building production systems. The world of LLM is relatively new, but the fundamentals of engineering and how you productionize a system have not changed from that perspective. Our clients bring in that expertise, that proven track record. And because we're building up the capabilities and experience in working with these open source models and the proprietary models, clients get all that when engaging with a partner like Nearform. You're not having to figure it all out yourself.

SE: The bulk of my work at Nearform has been building applications with open source tooling, specifically React and Next.js. They are wildly popular open source projects, and we use them on our Puma engagement, where Puma had a web presence that spanned many regions, close to 50 different geographies worldwide. Each region used Salesforce to manage and display their web content, which could get really unruly quickly. There was a lot of duplication, and a lot of inefficiencies. We were tasked with helping them transition to a per region platform and help them build a unified e-commerce platform, without sacrificing each location's unique requirements.

This was a really complex problem, where some regions have some specific requirements - some customization was needed, but also we wanted to unify a lot of the duplicative work that was being done - so that teams could evolve and iterate faster, as well as keep a level of consistency across the branding and messaging globally.

This meant we needed to allow individual locations to manage their content languages, payment processors, and then also use other third-party vendors while still enabling each site to scale and change independently. React gave us a powerful foundation for building a reusable component library that allowed the team to create consistent experiences across the sites. And the application itself is a server-rendered Next.js site - and for an e-commerce website, server rendering is crucial. It allows for rapid page loads, high cacheability, and consistent SEO performance. Those two contributed to a great new developer experience. It's supported worldwide by open source developers.

CH: Nearform has that history of leveraging open source from building solutions for clients - we've done this from day one. From day one, we've been an open source contributor, and an open source consumer. We really see the value in that model and the benefits it brings. Because you have visibility into what you're building, and how it was built, you can see all the way down the stack.

For example, we've worked with the HSE (Health Service Executive) in Ireland for the last year and a half on the HSE health app. From a technology perspective, you know that was leveraging a lot of open source tooling. So you had React Native from a front end perspective, you've got Node.js in the back end, and lots of other pieces of the system that are all open source tools.

I think the stat is 96% of all software and is now leveraging open source. So even if it's a closed source system, it's built using open source, right? It's almost impossible to build software now without leveraging open source in some manner.

Why? Because your alternative to not leveraging open source is you build everything yourself from scratch. So if you're trying to stand up a solution, you have to build every single piece of it yourself. The overhead is too large. The whole point of the open source system is that it allows you to accelerate and you can focus on delivering product value as opposed to spending a lot of time and energy on some of the underlying pieces of a system that are needed, but they're not moving you forward from an actual product perspective. Every solution we do is leveraging open source.