Elise AI automates leasing and maintenance with hybrid model stack
Elise AI has fully automated apartment leasing workflows and a significant portion of maintenance requests, handling millions of interactions monthly. The company relies on a hybrid technical approach, combining large language models with traditional machine learning to ensure reliability in essential industries.
Elise AI, a developer of conversational AI for essential sectors such as housing and healthcare, has implemented full automation for the apartment leasing process. The system allows prospective residents to book tours, receive access codes, and tour units without human intervention, streamlining the journey from initial website interest to move-in. The company also automates a substantial share of resident maintenance requests, where AI determines urgency, escalates issues, identifies vendors, and schedules physical repairs, though the final repair work still requires human labor. These systems process approximately 5 million phone calls per month, totaling about 60 million annually, alongside hundreds of thousands of emails and text messages daily.
Mario Martone, a leader at Elise AI, argues that production AI in essential industries requires a mix of large language models and traditional machine learning to achieve necessary reliability. He notes that large language models alone often fail on deterministic and probabilistic tasks at scale, such as predicting lease renewal risks, where the company uses in-house neural networks or traditional machine learning instead. This hybrid stack is critical for maintaining accuracy across the company's high-volume operations.
To maintain performance, Elise AI's applied research team employs a closed-loop experimentation process rather than relying on external benchmarks or public claims for new speech-to-text or text-to-speech models. When new models are released, the team immediately tests them against proprietary benchmarks and the current production stack to identify specific bottlenecks in their voice latency architecture, which performs tens to hundreds of operations within a 1.5-second window. The team also focuses on fine-tuning open-source models, leveraging the research backgrounds of staff, including theoretical physicists, to generate new ideas for model training. This strategy allows for rapid integration of improvements, ensuring the AI systems remain optimized for low latency and high accuracy. Elise AI is currently expanding into healthcare and focusing on reducing operational costs in affordable housing, a sector with minimal existing technology depth.