Mario Claudio Martone, Head of Applied Research at EliseAI, describes moving from theoretical physics into AI systems whose decisions affect housing and healthcare. He argues that the task should determine the model: language models are useful for conversation, while renewal prediction can benefit from more traditional machine learning and purpose-built networks.
The interview explains why small demonstrations cannot establish production reliability. The team combines researchers and engineers to evaluate large volumes of conversations, inspect consequential mistakes and compare candidate voice components on internal benchmarks rather than accepting external benchmark claims. Owning the stack helps the team locate latency bottlenecks and test alternatives in a short feedback loop.
The discussion connects these methods to leasing, tour access and resident maintenance requests. Digital steps can be automated while physical repairs still require people. The research priorities include anticipating new model capabilities, monitoring autonomous conversations at scale and reducing administrative costs in affordable housing; these are company goals and reported experiences, not guarantees of error-free service.
Watch on YouTube
