Spaces:
Sleeping
title: GroundTruth
emoji: 🏠
colorFrom: blue
colorTo: blue
sdk: docker
sdk_version: 5.16.0
app_port: 7860
app_file: app.py
pinned: true
license: apache-2.0
tags:
- real-estate
- spatial-reasoning
- robotics-ai
- gemini-api
- proptech
GroundTruth: Temporal Property Sentinel
A High-Fidelity Spatial Reasoning Engine for Real Estate Analysis
Project Overview
GroundTruth is an analytical platform designed to provide "Spatial Truth" in property valuation and assessment. By utilizing the Google Gemini Robotics-ER 1.5 model, this application moves beyond standard object detection to perform forensic, multi-period structural audits.
Why Robotics-ER 1.5?
Standard Vision-Language Models (VLMs) are trained to identify what is in an image (e.g., "a house with a lawn"). GroundTruth uses the Robotics-ER 1.5 API because it is specifically optimized for Embodied Reasoning—the ability to understand physical structure, depth, and spatial relationships as they unfold across time.
Key Technical Advantages:
- Spatial Accuracy: Outperforms standard models (like Gemini Flash) in 3D detection and precise pointing tasks, which is critical for identifying specific structural defects like roof sagging or foundation cracks.
- Temporal Reasoning: Natively understands cause-and-effect relationships and sequences of events. It doesn't just see two photos; it reasons about the maintenance trajectory between them.
- Physical World Logic: Trained on robotic interaction data, the model understands "affordances"—the physical possibilities of a space (e.g., whether a wall is likely load-bearing based on its position).
Features
- Side-by-Side Forensic Audit: Compare historical and present-day imagery to identify capital improvements or systemic neglect.
- Maintenance Trajectory Scoring: Automated classification of a property as "Improving," "Stable," or "Declining" based on visual structural evidence.
- High-Budget Thinking: Utilizes the model's tunable thinking budget to prioritize forensic accuracy over simple latency for complex structural questions.
Intended Use
- Real Estate Agents: Automating "Pride of Ownership" reports for listing presentations.
- Investors: Remote due diligence and asset condition monitoring over time.
- Compliance Officers: Identifying unauthorized additions or zoning violations via historical imagery comparison.
Limitations & Disclaimers
- Vision-Only: Analysis is based on exterior visual data; it cannot "see" internal structural integrity or hidden plumbing/electrical issues.
- Safety: While optimized for physical reasoning, all AI outputs should be verified by a licensed human inspector before financial or safety decisions are made.
License
This project is licensed under the Apache License 2.0, providing explicit patent protection and commercial flexibility for the Real Estate tech ecosystem.
Created by Evan Bench — Google AI Architect