Skip to content
<Hyun/>
All projects

StormVision-3D

Capstone · Research explanation

Turn storm data into an explanation.

Trace the documented research pipeline while keeping observed evidence, generated learning material and future work separate.

Interactive explanation · not the live service

Loading interactive explanation…

Explore the architectureSystem diagrams, design decisions and detailed flows

Documented AI pipeline

  1. NOAA · NASA · FEMAlive + historical data
    Step 1, storm records
    AI / ML PROCESSINGNeo4j Graph-RAGNLP / LLM retrieval
  2. NOAA · NASA · FEMA
    Step 2, satellite imagery
    AI / ML PROCESSINGStable DiffusionLoRA + ControlNet
    frames
    AI / ML PROCESSINGVideo generationStable Video Diffusion
    video clips
    ACCESSIBILITY & UX3D storm viewertimeline + narrate
  3. Neo4j Graph-RAG
    context
    ACCESSIBILITY & UXAI narration + TTSvoice explanations
    Step 3, narration
    3D storm viewer

Other connections

  • NOAA · NASA · FEMAtoDashboarddata feed
The pipeline from the team's poster: public hurricane data feeds (1) Graph-RAG retrieval and (2) generative image and video models, and (3) learners see the results with AI narration in a 3D viewer, next to a dashboard. Dashed links are inferred from the poster rather than drawn on it.
View the diagram as text

Parts

  • NOAA · NASA · FEMA (live + historical data): External service
  • Dashboard (Hurricane Irma case study): Client / UI
  • Neo4j Graph-RAG (NLP / LLM retrieval): Data store / broker
  • Stable Diffusion (LoRA + ControlNet): Owned service
  • Video generation (Stable Video Diffusion): Owned service
  • AI narration + TTS (voice explanations): Owned service
  • 3D storm viewer (timeline + narrate): Client / UI

Connections

  • Step 1: NOAA · NASA · FEMA → Neo4j Graph-RAG: storm records
  • Step 2: NOAA · NASA · FEMA → Stable Diffusion: satellite imagery
  • Stable Diffusion → Video generation: frames
  • Video generation → 3D storm viewer: video clips
  • Neo4j Graph-RAG → AI narration + TTS: context
  • Step 3: AI narration + TTS → 3D storm viewer: narration
  • NOAA · NASA · FEMA → Dashboard: data feed

How a hurricane lesson is produced

  1. Collect public data

    NOAA hurricane data, NASA satellite imagery and FEMA disaster declarations form the data layer, together with historical storm records such as Hurricane Irma (2017).

  2. Retrieve grounded facts

    An NLP/LLM layer uses a Graph-RAG framework on Neo4j to retrieve context for explanations.

  3. Generate storm imagery

    LoRA-tuned Stable Diffusion with ControlNet produces realistic hurricane visuals from the scientific and public data.

  4. Animate into video

    Stable Video Diffusion reconstructs the scenes as video, which a 3D reconstruction viewer plays back on a timeline.

  5. Narrate and explain

    AI narration with text-to-speech explains how the storm forms, moves and causes damage, while the dashboard and case-study modules present the live and historical data.

The project & my contribution

Problem
Hurricane data from NOAA, NASA and FEMA is hard for a general audience to follow.
Solution
An AI pipeline that retrieves facts with Graph-RAG and generates narrated, explainable 3D visualizations, shown in a Hurricane Irma case study.
Result
Placed 3rd in the AI Agent Track at Hack-A-Roo.
  • 3rd place, Hack-A-Roo AI Agent Track
Explore featuresVideo(opens in a new tab)Paper(opens in a new tab)Poster(opens in a new tab)

Pipeline

  • Graph-RAG retrieval on Neo4j, with a self-reported 94% retrieval accuracy
  • LoRA-tuned Stable Diffusion with ControlNet, and Stable Video Diffusion for video
  • 3D simulator, dashboard and case study, with AI narration and text-to-speech

Notes

  • Based on the team's poster and architecture graphic, not source code
  • A forecasting agent and multilingual narration are listed as future work