---
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*