** Bolt across the world with AI-powered geolocation
** Hack the Valley X Winner - Best UI Hack
Ever see a photo and think, "This place looks incredible. Where is that?"
Traditional reverse image search tools rarely work well for geolocation.
That’s why we built rainbolt.ai: an AI-powered platform that teaches you how to identify real-world locations from any image, providing context, culture, and insight.
rainbolt.ai combines AI, reasoning models, and geographic data to pinpoint coordinates, display nearby street views, and explain its thought process — analyzing road signs, vegetation, architecture, and more.
This project is inspired by Trevor Rainbolt, the world’s most famous geoguesser.
Core Capabilities
- Image Geolocation: Upload any photo to get precise coordinates.
- AI Reasoning: Watch the system explain how it identified features.
- Street View Integration: Explore nearby Mapillary street views.
- Interactive Chat: Refine or regenerate predictions interactively.
- Cultural Insights: Learn about landmarks, culture, and geography.
Technical Features
- Real-time Streaming using WebSockets for live reasoning updates.
- Multi-modal Analysis combining visual, textual, and spatial data.
- Confidence Scoring with transparent accuracy metrics.
- Dynamic Visualization via interactive globe and constellation views.
rainbolt.ai connects multiple systems into a single retrieval-augmented pipeline:
User Image → CLIP Encoding → Pinecone Vector Search → Gemini Reasoning → Mapillary Verification → Cultural Context
(Add a system diagram here, e.g., docs/images/system-diagram.png)
| Category | Technologies |
|---|---|
| AI & ML | Gemini API, OpenAI CLIP, LangChain, Pillow |
| Backend | FastAPI, Python, WebSockets |
| Frontend | Next.js, React, TypeScript, Tailwind CSS |
| Database | Pinecone (Vector DB), Firebase (Realtime) |
| Infrastructure | Google Cloud, Auth0, Mapillary API |
| State Management | Zustand, Radix UI |
- Image Processing – CLIP encodes visual features into vector embeddings.
- Vector Search – Pinecone retrieves similar geographic embeddings.
- AI Reasoning – Gemini analyzes matches and approximates coordinates.
- Verification – Mapillary street views validate and refine predictions.
- Presentation – Frontend displays cultural insights and dynamic visualizations.
- YFCC100M dataset for embeddings
- Geoguessr community guides for geographic hints
- Mapillary API for street-level imagery
- Global geographic and cultural datasets for context
- Hack the Valley X (2025)
- Winner: Best UI Hack (Sponsored by Conrad Mo)
- Built in 36 hours by a 4-person team
- Integrated Three.js for smooth globe animation and camera rotation.
- Deployed the complete AI pipeline on Google Cloud.
- Implemented real-time street-level vision via Mapillary API.
- Tuned Gemini reasoning prompts for consistent accuracy.
- Video Analysis: Extend CLIP to analyze video frames sequentially.
- Real-time Tracking: Monitor geographic changes over time.
- 3D Spatial Mapping: Generate depth maps and 3D reconstructions.
- Cultural Enrichment: Add deeper historical and sociological insights.
- Mobile App: Native app experience for on-the-go geolocation.
| Member |
|---|
| Daniel Pu |
| Daniel Liu |
| Evan |
| Justin Wang |
- Devpost Submission: https://devpost.com/software/rainbolt-ai
Proprietary. Built for Hack the Valley X 2025.