Geospatial Physical AI for disaster evidence, resilience intelligence, and auditable property decisions.
Rayford AI builds Ray, a geospatial AI system that connects remote sensing, street-level imagery, hazard context, and multimodal reasoning into property-level evidence for disaster assessment, claims triage, and resilience planning.
The organization is now structured as a research-to-product workspace: public repositories show the scientific foundation, private advisor-access repositories hold active disaster datasets and experiments, and product workspaces translate the research into Ray workflows.
Current organization view, as of July 11, 2026:
| Area | Status |
|---|---|
| Public surface | Organization profile, website, and selected research repositories. |
| Advisor-access research | Private imports for datasets, cross-view reliability, preparedness, and disaster-agent pipelines. |
| Product and operations | Private Rayford AI company, product, and operating workspaces. |
| Main technical theme | Auditable disaster intelligence from street-view, satellite, aerial, and geospatial evidence. |
Public links work for anyone. Repositories marked access required are private in the Rayford AI organization and require collaborator access.
| Repository | Role |
|---|---|
| .github | Organization profile and shared public identity. |
| Rayford-AI.github.io | Public website for Rayford AI. |
| Bi-Temporal-StreetView | Hyperlocal hurricane damage assessment from paired pre/post street-view imagery. |
| DamageArbiter | CLIP-enhanced multimodal arbitration for street-view disaster damage assessment. |
| Sat2Street-DisasterGen | Generative satellite-to-street framework for post-disaster street-view synthesis. |
| RAPID | Reproducible multi-agent pipeline for interpretable disaster damage assessment from satellite and street-view imagery. |
| GenAI4Dresilience | Publication-safe companion for generative AI across the disaster-resilience lifecycle. |
| Repository | Role |
|---|---|
| disaster-crossview-datasets | Curated index for wildfire, hurricane, and paired cross-view disaster datasets; access required. |
| CrossViewGate | Visibility-conditioned reliability gating for street-view and overhead damage assessment; access required. |
| PrepStreet | Preparedness-oriented street-view analysis linked with hurricane forecasts, spatial context, and hazard features; access required. |
| DisasterPilot | Multi-agent event pipeline from public disaster watch to dossier, exposure, evidence, and decision products; access required. |
| rayford-ai | Company workspace, brand, website material, and product-roadmap notes; access required. |
| hermes-work-manual | Internal operating manual for company workflows; access required. |
The imported disaster projects now form a coherent stack rather than isolated repositories:
| Layer | Repositories | Research question |
|---|---|---|
| Data foundation | disaster-crossview-datasets, Bi-Temporal-StreetView | What imagery, metadata, labels, and provenance are available across wildfire, hurricane, and multi-hazard settings? |
| Street-level damage perception | Bi-Temporal-StreetView, DamageArbiter | How much does pre-event context improve property-level damage assessment, and how can visual-language signals make results interpretable? |
| Cross-view reliability | CrossViewGate, RAPID | When street-view and overhead evidence disagree, which view should be trusted and when should the system defer? |
| Generative evidence support | Sat2Street-DisasterGen, GenAI4Dresilience | Can generative models support scenario reasoning or missing-view analysis without being mistaken for observed evidence? |
| Preparedness modeling | PrepStreet | Can pre-disaster street-view, hazard forecasts, and spatial context anticipate likely post-event damage patterns? |
| Operational agents | DisasterPilot, RAPID | How can public event monitoring, exposure modeling, evidence assembly, and decision support become an auditable pipeline? |
For a fast review:
- Start with disaster-crossview-datasets to understand the data inventory, storage policy, and disaster coverage.
- Read Bi-Temporal-StreetView and DamageArbiter for the street-view damage-assessment foundation.
- Read CrossViewGate for the current reliability problem: conflict-aware cross-view damage assessment.
- Read PrepStreet for the preparedness and pre-event modeling direction.
- Read DisasterPilot for the end-to-end event workflow from public watch to decision products.
- Use RAPID, Sat2Street-DisasterGen, and GenAI4Dresilience as the broader GenAI and multi-agent research context.
Rayford AI's disaster work is guided by a few non-negotiables:
- Auditable outputs: Every score should point back to imagery, metadata, source files, assumptions, and confidence.
- View-aware reasoning: Street-view, satellite, aerial, drone, and field imagery are not interchangeable; each view has its own failure modes.
- Fail-closed behavior: Missing imagery or missing metadata should produce a recorded evidence gap, not a fabricated answer.
- Human review: Ray should prioritize and explain evidence, not replace final claims, emergency-management, or recovery decisions.
- Data discipline: Raw imagery, protected datasets, and large generated artifacts stay private or local unless their license and review status allow public release.
Selected public research lines behind Ray include:
- Computers, Environment and Urban Systems 2025: Hyperlocal disaster damage assessment using bi-temporal street-view imagery.
- 2026 preprint: DamageArbiter, a CLIP-enhanced multimodal arbitration framework for hurricane damage assessment.
- IGARSS 2026: Satellite-to-Street, synthesizing post-disaster street views from satellite imagery via generative vision models.
- RAPID: A reproducible multi-agent pipeline for interpretable disaster damage assessment from satellite and street-view imagery.
- GenAI4Dresilience: Generative AI for mitigation, preparedness, response, and recovery.
Public research context is also available at AutoGeoAI4Sci.
| Member | Role |
|---|---|
| Yifan Yang | Founder and Technical Lead; multimodal spatial intelligence, street-view analysis, model arbitration, and autonomous GeoAI systems. |
| Dr. Lei Zou | Scientific and Technical Advisor; GeoAI, spatial intelligence, and disaster resilience direction at Texas A&M University. |
| Dr. Zhengzhong Tu | Technical Advisor; computer vision, multimodal model design, and validation strategy. |
| Dr. Heng Cai | Technical Advisor; built environment context, infrastructure intelligence, and product-risk review. |
Rayford AI separates public visibility from private product and research development.
- Public repositories may include company profiles, website assets, publication-safe research code, selected demos, and collaboration-facing summaries.
- Private repositories hold active experiments, dataset indexes, product code, customer discovery, pilot materials, and operating workflows.
- Public repository licenses apply only to the repository-authored materials they cover. Underlying third-party imagery, datasets, model outputs, and protected records remain subject to their own terms.
Rayford AI is not an open-source project by default. Public materials are shared for visibility, evaluation, advisor review, and research collaboration; proprietary product work remains protected unless explicitly released.
- Website: rayford-ai.com
- GitHub: Rayford-AI
- LinkedIn: Rayford AI
- Contact: contact@rayford-ai.com
Every property, ready and recoverable.
