From theory to edge AI: Ben Chang on his summer at Aeris-UK

Cambridge engineer Ben Chang talks about edge AI, client-facing R&D and the experience of turning technical ideas into working systems.

Ben Chang joined Aeris-UK in summer 2026 as a Machine Learning Research Engineer Intern. Over three months, he contributed to simulation, edge AI for security and synthetic data generation, seeing ideas develop from early-stage exploration through to demonstrations with clients. Before returning to Cambridge for an MPhil in Advanced Computer Science, we spoke to him about what he had learnt and how the experience shaped his thinking about a career in AI.

You started at Cambridge studying Maths. Why did you switch to Engineering?

I realised I was more interested in building things than spending most of my time on abstract theoretical work, so after my first year I switched into Engineering. My previous internship had focused more on software development and algorithms. I hadn’t yet had the chance to use machine learning professionally, so when I saw the Aeris role on LinkedIn, it felt like a good fit.

What appealed to you about joining a smaller company?

I like the fact that, in a small company, even as an intern you can have quite a lot of responsibility. You can contribute to things that are important and directly useful to the business. It is also very different building something that will actually be used, rather than completing a university exercise or something purely experimental.

What have you been involved in at Aeris-UK?

I’ve worked across three main areas. The first was more focused on software engineering and simulation. The largest part of my time was then spent on edge AI for security: looking at whether a cyber-attack could be detected by a small device operating in the field. What I particularly liked was being involved throughout the process. I wrote scripts to simulate attacks, helped decide what data to use, built models and then helped get something running on a phone that we could demonstrate to the client. More recently, I’ve also been looking at synthetic data generation. The areas are quite different, which has been useful because I’ve had exposure to a broad range of technical problems rather than doing one narrow task throughout.

What did running edge AI on a device teach you?

There are a lot more constraints to think about. If a device has limited memory, you have to balance how good the model is against whether it will fit and how quickly it can respond. A larger model might be more accurate, but if it takes too long to produce an answer, that can make it less useful. In situations where you need to react quickly, latency matters. Those trade-offs become much clearer once you start testing them in practice rather than only learning about them academically.

As part of his internship, Ben joined the Aeris-UK team at the Applied Machine Learning for Cyber Security (AMLUCS) event, gaining experience of an industry event where Aeris-UK both exhibited and presented its work.

Was that the most interesting part of your time with us?

Yes, probably. It was also the most challenging. There was a lot to deliver in a relatively short period, alongside regular updates, presentations and meetings with the client. It was fast-paced, but not stressful. It was a good pace. It was also my first experience of being directly involved in client discussions. In my previous role, I was contributing to commercial products, but most of my day-to-day interaction stayed within the company. Here, I was part of conversations about progress, feasibility and different technical approaches. That was completely new to me.

What did you learn from dealing directly with clients?

Sometimes the client had a more ambitious idea of what could be achieved than we initially thought was feasible, so there were times when we had to push back. At the same time, those expectations pushed us further. I think we were all surprised by how much we managed to get done by the end. The relationship was positive. If we disagreed with something or suggested another approach, they were willing to listen and take that on board.

Did anything surprise you about working with data?

Before I had worked with data in this kind of setting, I thought cleaning it would be fairly straightforward. There are standard techniques, so I assumed it would mostly be a case of applying them. Once you start looking at the data you have actually been given, though, all sorts of unexpected issues appear. Some things may be broken, information may be missing or you may not know exactly where something has come from. Figuring out how to handle that properly can require much more thought than I expected. It turned out to be more challenging, but also more interesting.

Aeris-UK is largely remote. What has the team environment been like?

I speak to the rest of the team quite frequently, probably several times a day through Slack or Notion. I also have regular catch-ups with my line manager and short daily meetings with the people involved in the same piece of work. Overall, it’s a very friendly place and people are easy to approach. One thing I’ve particularly picked up is the discipline around documentation. Everyone keeps things updated on Notion or GitHub so that other people know what is happening. I wasn’t as consistent about that before I joined, so seeing how the team operates has definitely improved my own habits. The daily catch-ups help as well. There is time set aside to raise problems, discuss ideas and make sure everyone stays up to date.

Has the experience changed what you want to do next?

More than anything, it has confirmed that this is an area I want to pursue. Before you have actually spent time in a field, you do not really know whether you like the idea of it or whether you will enjoy the day-to-day reality. Having done it now, I know it is something I want to continue with. The internship has also introduced me to agentic AI, which is an area I would like to explore further. After my MPhil, I am considering a PhD in AI, possibly in reinforcement learning. The experience has helped me narrow down what interests me and introduced me to areas I probably would not have looked at as closely before.