Draw a box on a map. Get back where to dig a village pond, the ground that drains into it, and how much water it'll hold in an average year — all overlaid on the map.
- Pick the land — search a village, then drag a box or draw a polygon around it.
- Free elevation — SRTM terrain tiles fill in the heights, no upload needed. Have a surveyed contour sheet instead? Upload the KML/KMZ for the same answer at survey accuracy.
- Finds the pond site — routes water downhill across the terrain, ranks locations by how much drains into them, and keeps a safe distance from watercourses.
- Estimates the water — ten years of real rainfall for that spot, run through a standard runoff model, turned into a fill curve for the pond.
- Shows the work — catchment boundary, flow paths, contours, and (from satellite imagery) existing ponds and buildable land, all on the map. Export as GeoJSON or PNG, save an analysis, reopen it later.
A sample area near Raipur, Chhattisgarh is one click away and resolves in about half a second.
docker compose up -d --build
# open http://localhost:5229 and press "Try the sample area"or hit the API directly:
curl -X POST http://localhost:5229/api/v1/analyzeArea -H 'Content-Type: application/json' \
-d '{"bbox": [81.2814, 21.2398, 81.3126, 21.2636]}'Full setup instructions, including a single-process dev server, are in docs/INSTALL.md.
browser ── frontend (nginx) ── gateway ─┬─ worker ─┐
├─ worker ─┼── cv (imagery)
└─ worker ─┘
saved sites: SQLite on the gateway's volume
One codebase, one Docker image — whether a process is a gateway, a worker, or the imagery service is decided entirely by its environment variables. Run it with nothing set and it just analyzes itself, which is what the live demo does.
| Path | What lives there |
|---|---|
app/core/ |
The analysis: DEM building, hydrology, catchment delineation, pond siting, contouring |
app/providers/ |
Elevation tiles and rainfall, each behind a cached interface |
app/cv/ |
Satellite imagery — existing water bodies and buildable land |
app/pipeline.py |
Ties it together: a contour sheet or a map area, same answer out |
app/jobs.py |
Async jobs with per-stage progress, a result cache, and dispatch across workers |
static/index.html |
The whole front end — one file, no build step |
docs/ |
API reference, methodology, validation against a real survey, and the load-test results |
Full endpoint reference: docs/API.md, or /docs on any running instance.
The free SRTM elevation was checked against a real 1 m contour survey of the same ground (docs/VALIDATION.md): the two DEMs agree to 0.26 m RMSE, and both recommend a pond in the same valley, 400 m apart. Good enough to find the site — for construction, still survey it.
pytest588 tests, no network required — elevation and rainfall are stubbed with known answers, so a run takes about three minutes and reproduces the same result every time.
- docs/REPORT_PHASE3.md — the written report for this system, or docs/report/ for the full technical report as LaTeX source and PDF
- docs/METHODOLOGY.md — the hydrology and siting method
- docs/SCALING.md — throughput and memory under load