EstateOS

Client Request for AI Agent Conversation
Portfolio
Intelligence Layer
Client-side AI Agent Conversation
3d Visualization
Interactive Map

Inspiration

The idea for EstateOS started when Ananth's cousin was house-hunting and told him how impossible it was to keep track of everything. After visiting five houses in one day, the details—like which one had the renovated kitchen or the better natural light—started to blur together. Real estate agents face the same struggle; they're juggling dozens of clients and can't remember every specific preference mentioned during a walkthrough. We realized there was massive friction because the most important data—the actual conversation—is usually lost the second everyone leaves the property. We wanted to build the "connective tissue" that captures that live intelligence and turns it into something useful for both sides.

What it does

EstateOS is a voice-driven AI ecosystem that digitizes the entire real estate experience. For agents, it’s a command center featuring an interactive Mapbox overview and an "Intelligence Page" that builds a live Knowledge Graph from their conversations. Whether using the app’s voice input or Meta Ray-Ban glasses, the AI extracts entities like buyer preferences and property features in real-time. Agents can also instantly generate 4K technical schematics and personalized PDF reports through a chat interface, which are then emailed directly to clients via Composio.

For clients, the platform offers a "Client Portal" that goes way beyond static photos. They can view AI-generated ads via Google Veo, explore their interests through 3D visualizations, and even enter an interactive simulation to change a room's color scheme or layout in real-time. To solve the "always-on" support problem, clients can jump into a real-time AI voice call for general property details, ensuring they get answers even when their agent is busy.

How we built it

We built the core of EstateOS using Next.js 16 and React 19 for a fast, multimodal frontend, backed by a FastAPI and Node.js server.

Spatial Data: We used Mapbox GL for property tracking and react-force-graph-3d to visualize the relational Knowledge Graph.

The Intelligence Layer: We integrated Google Gemini and OpenAI for entity extraction, while Vapi and LiveKit handled the low-latency voice orchestration.

Multimodal Assets: We used Google Veo for video ads, Odyssey.ml for the interactive simulations, and Hunyuan World for 3D Gaussian Splatting.

Automation: Composio handled the agentic email routing, and PDFKit allowed us to turn chat prompts into professional property reports on the fly.

Challenges we ran into

Our biggest hurdle was the hardware integration. We initially planned to rely heavily on Meta Ray-Bans, but the VisionClaw bridge proved to be pretty unstable during the hackathon. We were able to livestream and record, but getting the live voice input to be precise required a lot of pivoting. We eventually leaned into a more robust voice-over-WebRTC approach for the "Intelligence" page to ensure the data was accurate. Merging so many disparate AI libraries (Veo, Odyssey, Mapbox, and the Knowledge Graph) into one cohesive dashboard also led to some "dependency hell," but we managed to get them all talking to each other through a shared data ecosystem.

Accomplishments that we're proud of

We’re incredibly proud of building such a complex, end-to-end system in just a weekend. It wasn't just about making a "wrapper"—we actually built a pipeline where a spoken sentence can trigger a knowledge graph update, which then informs a generated PDF report, which is then emailed to a client automatically. Integrating the Gaussian Splatting and interactive 3D simulations also felt like a huge win, as it moves the needle from just "looking at a house" to actually "experiencing" it.

What we learned

We learned a lot about Spatial Computing and the difficulty of processing live, unstructured audio into structured data. We also got a crash course in "Agentic Tool Use"—learning how to let an LLM decide when to generate a schematic versus when to send an email via Composio. Most importantly, we learned that in a high-stakes industry like real estate, the UI/UX has to be invisible; the agent needs to be able to talk naturally while the AI does the heavy lifting in the background.

What's next for EstateOS

The next step for EstateOS is hardening the security by adding full authentication for separate client and realtor dashboards. We also want to train our reporting engine on way more local data—integrating Maps APIs for distance-to-amenities, school district data, and neighborhood crime stats so the AI can answer hyper-specific questions about "things nearby."

We’re also looking to make the voice integration much more robust. The goal is to move toward a truly hands-free, end-to-end experience where every single feature—from generating a 3D render to sending a complex report—is entirely voice-controlled with quicker, more dynamic changes. Finally, we want to make the Meta glasses integration more precise and stable, so the agent never has to look at a screen while they’re demoing a home.