RESI Inspections
AI-powered property inspections that adjust appraisals based on visual condition analysis.
RESI Inspections is currently under development. This page describes planned functionality.
Overview
RESI Inspections extends the appraisal system by analyzing property condition through uploaded images. AI models assess visible condition factors. This inspection data adjusts base appraisal prices to reflect actual property state.
How inspections work
Step 1: Property owner uploads images
Property owners or inspectors upload photos of the property including exterior views, interior rooms, kitchen, bathrooms, mechanical systems, and structural elements.
Step 2: AI models analyze images
Miners submit AI models trained to assess property condition from images. Models evaluate factors including structural integrity, maintenance level, renovation quality, visible damage, system age and condition, and aesthetic updates.
Step 3: Condition scoring
Models generate condition scores for different property aspects. Scores aggregate into overall property condition rating. Condition rating translates to price adjustment factor.
Step 4: Price adjustment
Base appraisal price from ML competition receives condition adjustment. Final price reflects both market data and actual property state.
Example: Base appraisal of $500,000 for property. Inspection reveals excellent condition with recent updates. Condition adjustment factor of +5%. Final valuation of $525,000.
Example: Base appraisal of $500,000 for property. Inspection reveals deferred maintenance and aging systems. Condition adjustment factor of -8%. Final valuation of $460,000.
Competition structure
Similar to appraisals, inspections operate as continuous ML competition on Bittensor Subnet 46.
Miners
Task: Submit AI models that accurately assess property condition from images.
Process:
- Train computer vision models on property images with known condition factors
- Upload models to Hugging Face
- Register model hash on-chain
- Earn rewards when condition assessments prove accurate
Training data sources: Public property listings with images and condition descriptions, inspection reports paired with photos, renovation before and after photos, and insurance claim photos with damage assessments.
Validators
Task: Benchmark submitted inspection models against ground truth data.
Process:
- Collect properties with known condition and recent sales
- Run inference on submitted inspection models using property images
- Calculate condition scores from model outputs
- Compare condition-adjusted prices to actual sale prices
- Set weights based on accuracy improvement
Key measurement: Models improve price prediction accuracy when condition adjustments are applied. Models that improve predictions receive higher weights.
Use cases
Enhanced RWA lending
Lenders request both appraisal and inspection. Base valuation from market data. Condition adjustment from visual inspection. More accurate collateral assessment reduces risk.
Property tokenization
Tokenized properties include verified condition data. Initial inspection establishes token backing. Periodic re-inspection detects condition changes. Transparent condition data builds investor confidence.
Insurance underwriting
Insurance protocols assess property risk using inspection data. Condition scoring identifies maintenance issues. Premium pricing reflects actual property state. Periodic re-inspection tracks risk changes.
Maintenance tracking
Property owners track condition over time. Regular inspections identify deterioration early. Maintenance interventions prevent value loss. Documentation supports resale value claims.
Renovation verification
Property improvements verified through before and after inspections. Condition score improvements quantify value added. Objective verification supports refinancing. Transparent documentation increases buyer confidence.
Technical approach
Image requirements
Minimum image set: Exterior front, rear, sides, roof, interior main rooms, kitchen, bathrooms, mechanical systems, and basement or foundation.
Image quality: Minimum resolution, adequate lighting, and clear focus on condition-relevant features.
Metadata: Capture timestamp, location verification, and image authenticity proof.
Model architecture
Models use computer vision techniques including convolutional neural networks for feature extraction, attention mechanisms for condition-relevant regions, multi-task learning for different condition aspects, and ensemble methods for robust predictions.
Condition categories
Models assess multiple condition dimensions including structural soundness, system functionality, aesthetic quality, maintenance level, and renovation recency.
Scoring methodology
Each condition category receives numerical score. Category scores weighted by impact on property value. Aggregate score translates to price adjustment factor. Adjustment factor applied to base appraisal.
Anti-gaming measures
Image authenticity: Verify images originate from claimed property and date.
Model diversity: Reward models using different approaches and architectures.
Consensus requirement: Multiple validators must agree on condition scores.
Ground truth validation: Regular benchmarking against properties with known condition and sale prices.
Integration with appraisals
Inspections complement appraisals to create complete pricing system.
Appraisal: Establishes base property value from market data including location, size, features, and recent comparable sales.
Inspection: Adjusts base value for actual condition including maintenance quality, system age, renovation level, and visible damage.
Combined output: Market-based valuation adjusted for property-specific condition factors.
API integration
Developers request both appraisal and inspection in single API call. Response includes base appraisal price, condition assessment scores, condition adjustment factor, and final adjusted price.
Example API response:
{
"base_appraisal": {
"predicted_price": 500000,
"model_id": "appraisal_model_123",
"r_squared": 0.89
},
"inspection": {
"condition_score": 8.5,
"condition_categories": {
"structural": 9.0,
"systems": 8.0,
"aesthetic": 8.5,
"maintenance": 8.5
},
"adjustment_factor": 1.05,
"model_id": "inspection_model_456"
},
"final_valuation": {
"adjusted_price": 525000,
"confidence_score": 0.94
}
}
Roadmap
Phase 1: Model development
Collect training datasets. Develop baseline inspection models. Establish condition scoring methodology. Create validation framework.
Phase 2: Competition launch
Release miner documentation. Begin accepting model submissions. Launch validator infrastructure. Start accuracy benchmarking.
Phase 3: API integration
Integrate inspection with appraisal API. Release developer documentation. Enable production API access. Launch SDK support.
Phase 4: Feature expansion
Add specialized inspection types including foundation, roof, mechanical systems, and environmental factors. Support additional property types. Enable time-series condition tracking.
Vision: Universal real world asset oracle
Inspections represent the second product in RESI's mission to become a universal oracle for real world assets on blockchain.
Current: Appraisals provide market-based pricing.
Next: Inspections add condition-based adjustments.
Future: Additional verification mechanisms for title verification, lien detection, permit compliance, environmental assessment, and expansion to other real world assets.
Goal: Complete, verifiable, decentralized pricing for any real world asset on blockchain.
Get involved
For miners
Start preparing computer vision models for property condition assessment. Monitor GitHub repository for competition launch announcements. Join Discord for technical discussions.
For validators
Review inspection competition design. Plan computational resources for image processing. Watch for validator software releases.
For developers
Understand inspection API design. Plan integration points in applications. Provide feedback on API structure.
For community
Follow development progress. Participate in design discussions. Share use case ideas.