Case study · Flagship product
Built end-to-end by Rocky Web Studio for Australian landholders, regen-ag consultants and mine-site rehabilitation teams.
The problem
A traditional keyline / permaculture plan takes a specialist days on the ground and hundreds of hours in QGIS. For Australia's millions of hectares of degraded pasture and post-mining land, that doesn't scale.
The approach
We ingest DEM and rainfall data for any selected parcel, derive flow accumulation and slope, and hand the result to a constrained LLM design pass. The output is a real plan a consultant can refine, not a chatbot answer.
Before / after
A degraded grazing block in Central Queensland, and the engine's fifteen-year design — contour swales, sited dams, multi-canopy food forest, riparian buffer, rotational cells.
Architecture
Next.js on Vercel for the front. TanStack-style server functions for typed RPC. Supabase (Postgres + Auth + RLS) for everything stateful. Google Maps + custom raster pipeline for geodata. An LLM design engine wrapped in evals. Resend for transactional email.
Key features
Click any Australian parcel — cadastre, contour and hydrology layers load instantly.
Keyline, swale, zone and species plans generated from terrain + climate.
Conversational AI that guides landholders through design decisions, species selection and climate adaptation — grounded in real parcel data.
Share-ready PDFs for landholders; GeoJSON / KML for the contractor.
Production auth, RLS-locked tables, email infrastructure done right.
Outcomes
We take on a small number of MVP and AI builds each quarter. If the problem is well-defined and the deadline is real, let's talk.