Earth Explorer (Desktop)

A desktop globe explorer using Electron + CesiumJS + FastAPI with:

• Real Earth imagery
• Continuous zoom from globe scale toward street scale
• Right-side chat panel for navigation commands (for example: show me Paris)
• Python backend query API
• Shape rendering for countries, rivers, and regions when geometry is available

Setup

From this directory:

npm run setup

This installs Node + Python dependencies, verifies required files, and checks runtime endpoints.

Run

npm start

Electron starts the FastAPI backend automatically with uv run (http://127.0.0.1:8000) and then launches the app window.

Commands

You can type:

• show me <place>
• go to <place>
• fly to <place>
• or just a place name

Chat requests go to backend query endpoint:

• POST /api/query

Example body:

{ "action": "geocode", "q": "France" }

Response includes point coordinates and, when available, GeoJSON geometry (Polygon, MultiPolygon, LineString, etc.). The frontend renders geometry overlays and flies camera to the shape extent.

Manual backend run (optional)

uv run --project backend uvicorn backend.main:app --host 127.0.0.1 --port 8000

Offline PDF Processing (Docling)

Standalone script:

uv run --project backend python backend/scripts/process_pdf_docling.py \
  --input /absolute/path/to/file.pdf \
  --output-dir backend/output

Output file format:

• backend/output/<input-stem>.docling.json

Offline Location Extraction (Docling JSON -> CSV via GLiNER2)

Standalone script:

uv run --project backend python backend/scripts/extract_locations_gliner2.py \
  --input-json backend/output/<input-stem>.docling.json \
  --output-csv backend/output/<input-stem>.locations.csv

Notes:

• First run downloads the GLiNER2 model from Hugging Face.
• Default model: fastino/gliner2-base-v1 (override with --model).
• Default labels include cities, countries, rivers, oil fields, and related location types.

Offline Location Extraction (Docling JSON -> CSV via NuExtract2)

Standalone script:

uv run --project backend python backend/scripts/extract_locations_nuextract2.py \
  --input-json backend/output/<input-stem>.docling.json \
  --output-csv backend/output/<input-stem>.nuextract.locations.csv

Notes:

• Default model: numind/NuExtract-2.0-2B (override with --model).
• First run downloads model weights from Hugging Face.

Offline Location Extraction (Docling JSON -> CSV via GPT-OSS on LM Studio)

Standalone script:

uv run --project backend python backend/scripts/extract_locations_gpt_oss_lmstudio.py \
  --input-json backend/output/<input-stem>.docling.json \
  --output-csv backend/output/<input-stem>.gptoss.locations.csv \
  --model openai/gpt-oss-20b \
  --base-url http://127.0.0.1:1234/v1

Notes:

• Ensure LM Studio is running with an OpenAI-compatible local server enabled.
• Ensure the selected GPT-OSS model is loaded in LM Studio.

Notes

• Street-level imagery requires network access for map tiles.
• If tile download is blocked, the app will still run but detailed imagery may not load.