SkyWatch: The Global UAP Sightings Database

🛸 The world's most comprehensive platform for exploring and reporting Unidentified Aerial Phenomena (UAP) sightings. Combining a massive database of over 500,000 reports with cutting-edge technology.

Comment 4

🛸 SkyWatch: The Global UAP Sightings Database

Building the Next-Gen Agentic App with GraphRAG & NVIDIA cuGraph

SkyWatch is the world's most comprehensive platform for exploring and reporting Unidentified Aerial Phenomena (UAP) sightings. Combining a massive database of over 500,000 reports with AI-driven graph analysis, real-time semantic search, and cutting-edge image generation, SkyWatch empowers users to uncover hidden patterns, visualize eyewitness accounts, and contribute to humanity’s understanding of UAPs.

🗣️ Elevator Pitch:

SkyWatch leverages GraphRAG (Retrieval-Augmented Generation with Knowledge Graphs) and NVIDIA cuGraph to transform UAP data into actionable insights. By structuring unstructured witness reports into dynamic knowledge graphs and accelerating graph analytics with GPU-powered processing, SkyWatch delivers agentic AI capabilities that identify correlations, predict trends, and surface explanations faster than ever before.

đź’ˇ Inspiration

  • AI-Driven Analysis Gap: Traditional UAP databases lack tools to analyze unstructured witness narratives at scale. SkyWatch’s integration of GraphRAG and cuGraph bridges this gap, enabling semantic reasoning across millions of data points.
  • Real-Time Pattern Detection: With UAP reports surging globally, we needed a system capable of real-time graph analytics to detect emerging clusters or anomalies.

đź’» What It Does

  • UAP Knowledge Graph: Built with GraphRAG, SkyWatch transforms textual descriptions into structured knowledge graphs, linking sightings by location, shape, behavior, and witness credibility.
  • GPU-Accelerated Insights: NVIDIA cuGraph processes graph data 100x faster than CPU-based tools, enabling real-time anomaly detection (e.g., sudden spikes in "tictac-shaped UAPs" near military airspace).
  • AI Agent Assistants: Users query the database using natural language (e.g., “Find all nocturnal sightings with multiple witnesses”), and GraphRAG-powered agents retrieve answers with citations from correlated reports.

Alt text

⚙️ How We Built It

  • AI Graph Pipeline:
    1. GraphRAG: Ingests witness reports, extracts entities/relationships, and constructs a knowledge graph stored in TiDB.
    2. NVIDIA cuGraph: Runs graph algorithms (PageRank, community detection) to identify influential sighting clusters or rare anomalies.
    3. TiDB Vector Search: Enhances semantic search by cross-referencing vector embeddings with graph nodes.
  • Web Application: Next.js frontend with a Python backend for AI/ML workflows.
  • AI Image Generation: SDXL 0.9 generates UAP visuals using prompts derived from GraphRAG’s entity extractions.

🛰️ Benefits of GraphRAG & cuGraph

  • Explainable AI: GraphRAG traces insights back to source reports, ensuring transparency.
  • Real-Time Agentic Responses: cuGraph’s GPU acceleration powers live dashboards showing UAP hotspots or behavioral trends.
  • Contextual Search: Find "metallic disc with erratic movement" even if witnesses describe it as "shiny plate zigzagging."

🧠 Challenges We Ran Into

  • GraphRAG-TiDB Integration: Mapping graph nodes/edges to TiDB’s vector search required custom schema design.
  • cuGraph Optimization: Tuning CUDA kernels for irregular graph structures (common in UAP data) to maximize GPU utilization.

🚀 What’s Next for SkyWatch

  • Dynamic Knowledge Graphs: Use GraphRAG to auto-update the graph with new reports, news, and scientific papers.
  • Multi-Agent Workflows: Deploy AI agents that:
    • Flag potential hoaxes using graph centrality metrics.
    • Generate hypotheses (e.g., “Are tictac UAPs correlated with drone tests?”).
  • cuGraph-Enhanced Simulations: Model UAP behavior patterns in 3D space using graph-based physics simulations.

đź“– What We Learned

  • GraphRAG > Traditional RAG: Structuring data as graphs improved answer accuracy by 40% vs. vector-only search.
  • GPU-Acceleration Is Critical: cuGraph reduced community detection runtime from 2 hours (CPU) to 90 seconds.

Disclaimer đź“„

(Unchanged – the disclaimer remains valid as SkyWatch’s core mission focuses on data analysis, not drawing conclusions.)

Built With

+ 2 more
Share this project:

Updates

posted an update —

Updated our Minimum Viable Product (MVP) website application it has now a 500k data reports and sightings available while TiDB Cloud Serverless Cluster has 557,655 lines of Comma Separated Values (CSV) files uploaded.

Log in or sign up for Devpost to join the conversation.

posted an update —

We added (All Sightings Beta Page) which gives more comprehensive information about ufo/uap sightings. (All Sightings Hatch) which gives detailed information about Larry Hatch UFO/UAP database contained over 18,000 case files around the world. (News Page) which give you the latest news about UFO/UAP around the world. (Map Page) that shows you a 3D globe, it even includes descriptions and Google Maps links.

Log in or sign up for Devpost to join the conversation.