slideGuess

Inspiration

Landslides present catastrophic, life-threatening risks to mountain, hill, and valley communities globally. Traditional early warning systems often rely on isolated physical sensors that fail to predict long-term multi-year probabilities or synthesize complex contextual geospatial attributes. Inspired by the urgent need for an intuitive, interactive, and AI-driven predictive mapping tool, slideGuess transforms raw meteorological, seismic, and geological metrics into actionable, immersive 3D intelligence that empowers communities and researchers to anticipate hazards before landslides strikes.

What It Does

• Predictive Landslide Analysis: Automatically computes next landslide probabilities, exact years of probable occurrence, specific slide types (e.g., rotational slides, debris flows), volumes ($m^3$), speeds ($m/s$), and main geometric dimensions.
• 3D Globe Visualization Interface: Features an auto-rotating 3D WebGL sphere embedded with interactive needles. Hovering over any needle expands gummy-styled animated containers displaying precise location details while seamlessly pausing globe rotation. Clicking a needle routes the user directly to a detailed 3D data table view.
• Distributed Agent Memory: Maintains robust analytical state, memory context, and vector-backed historical queries securely via CockroachDB and LangChain.
• Automated Cloud Reporting: Triggers serverless AWS Lambda functions to compile and store analysis results in structured CSV format directly onto Amazon S3 buckets.

How We Built It

• Backend & Routing: Developed a robust, production-grade Express.js server in TypeScript with dedicated router modules handling data ingestion, model inference requests, and database queries.
• Database & AI Integration: Connected to CockroachDB for durable, distributed SQL agentic memory and integrated Ollama's llama3.2 model locally via LangChain. Configured the Managed CockroachDB MCP Server for safe, streamlined database tool execution.
  
• Serverless Pipeline: Integrated AWS Lambda to export completed LLM analytical payloads directly into Amazon S3 storage as clean CSV files.
• Frontend Experience: Designed a clean Tailwind CSS interface featuring a custom 3D HTML5 canvas/WebGL gummy sphere simulation complete with smooth mouse-tracking event hooks and dynamic overlay cards.

Challenges We Ran Into

• Real-Time 3D State Management: Synchronizing real-time 3D globe rotation pauses and resumptions with intricate mouse-hover and click events across dynamic DOM overlays.
• Vector Performance Tuning: Tuning LangChain vector embeddings with CockroachDB's distributed architecture to ensure lightning-fast spatial queries against massive historical landslide catalogs.
• Asynchronous Handshakes: Managing secure, reliable asynchronous handshakes and permission flows between Node.js, local Ollama execution endpoints, and AWS S3 storage configurations.

Accomplishments That We're Proud Of

• Successfully deploying a fully functioning local LLM pipeline using Ollama (llama3.2) combined with a distributed SQL and vector database backend.
• Building an engaging, highly responsive 3D gummy-themed needle navigation interface that bridges smooth WebGL animations with detailed data matrices.
• Establishing a robust, automated cloud-native tracking pipeline via AWS Lambda and S3 CSV exports.

What We Learned

• The immense power of combining traditional relational SQL tables with vector-ready distributed databases like CockroachDB for scalable, fault-tolerant AI agent memory.
  
• Advanced techniques for balancing heavy WebGL 3D canvas animations with lightweight, performant UI state management using Tailwind CSS.

What's Next for slideGuess

• Integrating real-time IoT seismic, tilt-meter, and rainfall sensor feeds directly into the CockroachDB instance for live anomaly detection.
• Expanding the 3D globe visualization to support multi-layer regional drill-downs, historical timeline scrubbing, and augmented reality (AR) field-view integration for emergency responders.

Built With

• amazon-web-services
• aws-lambda-function
• cockroachdb
• express.js
• langchain
• mcp-server
• node.js
• ollama
• postgresql
• tailwindcss
• typescript