Spaces:
Sleeping
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* |