MINDS postgraduate researcher Yaseen M. Osman is working to make large language models faster and less demanding to run. His experience with Rosie and Reachy shows why involving the public can help researchers understand where technology succeeds, where it struggles and what people may need from it.

A request the researchers had not expected

When a visitor at the British Science Festival asked a robot to read the time from their watch, it did not give the right answer.

For MINDS postgraduate researcher Yaseen M. Osman, however, the question was valuable. It suggested a possible accessibility use for conversational robots that he had not previously considered.

This kind of unexpected exchange is one reason Yaseen M. Osman sees public engagement as part of the research process, rather than simply an opportunity to demonstrate a finished piece of technology.

“Involving the public also informs us whether what we’re doing is actually needed or not, and how can we adjust our trajectory.”

Yaseen M. Osman, MINDS postgraduate researcher

Meeting Rosie and Reachy

Yaseen M. Osman worked on Talking Robots and joined MINDS CDT Enterprise Fellow Tommy Tham to run the demonstration at the British Science Festival in September 2026.

The project allows visitors to speak with two robots, Rosie and Reachy. Although they use the same underlying type of technology, the robots have contrasting personalities. Rosie is presented as kinder, while Reachy is more direct. Ask both robots the same question and they may give different answers, accompanied by different movements and reactions.

Visitors asked the robots questions, invited them to dance and tested how they responded to one another. These playful interactions made it easier to start conversations about how large language models work, but they also exposed the technology’s practical limitations.

In a busy public space, background noise could make it difficult for the robots to identify a question. Visitors also had to wait for an answer while the system processed their words and generated a response.

From a public demonstration to a research question

Those limitations connect closely with Yaseen M. Osman’s doctoral research.

Large language models can require substantial computing resources. His work investigates how to make them more efficient without losing the performance that makes them useful. The longer-term aim is to reduce the barriers to running these models on personal devices, rather than depending entirely on remote data centres.

This has practical implications. A model that can run locally may require less communication with remote infrastructure and can keep more data on the device being used. It may also continue operating when a network connection is unavailable.

The Talking Robots demonstration brings some of these questions into view. Faster responses could make a conversation feel more natural. Lower hardware requirements could make systems easier to deploy. Moving between software and a physical robot adds a further set of challenges involving speech, movement and the surrounding environment.

For Yaseen M. Osman, the work creates a connection between research on the efficiency of large language models and the experience of somebody trying to use an AI system in front of them.

Letting people test the technology

A controlled laboratory test can show whether a system performs a particular task. A public demonstration introduces different voices, expectations, surroundings and questions.

At the Festival, visitors noticed the response delay and the difficulty of filtering background noise. They also tried requests the researchers had not anticipated, including asking a robot to read the time from a watch.

These interactions do not establish that the robots are ready for accessibility or assistive applications. They do, however, reveal what people might want from future systems and where the underlying technology would have to improve.

The demonstration also gave visitors an opportunity to question the researchers directly. Conversations covered where the language model was running, the difference between local and cloud deployment, and why an organisation handling sensitive information might want to keep an AI system on its own infrastructure.

Giving research a public setting

Talking Robots is one part of Yaseen M. Osman’s wider public-engagement work.

Through Talking Itchen, he helped give the River Itchen an AI-generated voice linked to water-quality information. He also taught children from the Young Guardians of the River Itchen about large language models, beginning with an imaginary model in one dimension before building towards the thousands of dimensions used by real systems.

Across both projects, the technology becomes a starting point for conversation. Visitors can test it, question its answers and see its limitations instead of encountering AI only as an abstract idea.

That exchange also benefits the researcher. It provides direct evidence of what people find useful, confusing, frustrating or unexpectedly interesting.

Building AI around real needs

Talking Robots does not present conversational AI as a finished solution. Its value lies partly in showing what the technology cannot yet do reliably.

By placing the demonstration in front of a public audience, Yaseen M. Osman and Tommy Tham gave visitors an opportunity to explore the system for themselves. In return, the researchers heard questions and requests that could not have been completely planned in advance.

For Yaseen M. Osman, that process connects the visible experience of talking to a robot with the deeper technical challenge at the centre of his PhD: making large language models efficient enough to become faster, more practical and easier to deploy.

Watch the Talking Robots film to hear Yaseen M. Osman explain the project, or visit his MINDS profile to learn more about his research into smaller, greener and more private AI.