Inspiration The real estate industry moves incredibly fast, but the technology agents use often slows them down with cognitive overload. We noticed that while the Lofty platform has powerful AI capabilities, they are treated as add-ons rather than the main focus. Agents are still forced to manually scan multiple widgets just to figure out what they need to do first. We wanted to fix this by building a copilot that acts as a proactive operating system, anticipating the agent's needs so they can spend less time scrolling and more time closing deals.
What it does Lofty AOS completely transforms the daily dashboard. When an agent logs in, they receive a Morning Briefing that summarizes overnight activity and tells them exactly what to focus on for the day. It features a Live Activity Feed that watches buyer behavior in real-time. If a buyer shows high interest, the system automatically drafts personalized text messages and invites them to property showings. It also includes an AI Command Center that agents can talk or type to, allowing them to navigate the platform, summarize market data, and execute tasks completely hands-free.
How we built it We built the frontend using React, Vite, and Tailwind CSS for a fast, responsive user interface. To make the data real-time, we integrated Supabase, which updates the dashboard instantly whenever a buyer interacts with a property. For the artificial intelligence, we used a specialized hybrid approach. We used Llama 3.3 hosted on Groq Cloud to power the conversational AI Command Center because it offers massive speed and sub-second latency. We then deployed Anthropic's Claude 3.5 Sonnet to handle the delicate job of writing warm, human-sounding text messages to clients. We also created custom logic that scores leads based on how many times they view a property and how recently they visited.
Challenges we ran into Our biggest hurdle was latency. If an agent asks the AI to open a page or check a task, waiting even three seconds ruins the experience. We had to heavily optimize our API calls and switch to Groq's high-speed infrastructure to ensure the AI Command Center felt instantaneous. Also, connecting live database updates to our React components without causing the whole page to constantly refresh or crash required a lot of careful engineering. Finally, instructing Claude to write text messages that sounded like a real agent instead of a corporate robot took extensive prompt tuning.
Accomplishments that we're proud of We are incredibly proud of achieving a true sub-second conversational interface that actually controls the software. We successfully integrated two different powerful AI models smoothly into one cohesive product. We also managed to build a functional voice assistant, which is a massive quality-of-life improvement for agents who spend most of their days driving from property to property.
What we learned We learned the profound difference between simply adding an AI chatbot to a website and building an AI-native product where the AI acts as the central routing system. We discovered that routing different tasks to specialized AI models is much more efficient than relying on just one. Most importantly, we learned that the best user interface is proactive. By building the Morning Briefing to surface what matters immediately, we realized users value software that predicts their needs over software that gives them a hundred buttons to click.
What's next for Lofty AOS Next, we want to expand the AI Command Center's capabilities to handle full, autonomous transaction management from the initial offer all the way to the final closing paperwork. We also plan to integrate deeper market forecasting tools to help agents predict property trends in their specific zip codes, ultimately setting up a beta launch to get this tool into the hands of top-producing real estate teams.
Built With
- anthropic
- groq-cloud
- javascript
- llama-3.3
- postgresql
- react
- supabase
- tailwind-css
- vite
Log in or sign up for Devpost to join the conversation.