convez.ai
MCP ServerDados & AnalyticsAvançado
galois-dh

Heavi — Plataforma de Computação Espacial para Energia Renovável

galois-dh/abelian

Scoring determinístico de suitabilidade solar com auditoria completa de dados

Ver no GitHub
InicialPluginScore · de 0 a 100

MCP server que fornece ferramentas de análise espacial determinística para avaliação de aptidão de terrenos em energia solar, análise de risco (incêndio/inundação) e estudos de trade-area. Integra 15 fontes de dados federais e abertos, com metodologia auditável e pesos calibrados por região NERC contra 6.321 instalações solares operacionais.

Instalação

$npx convez-ai galois-dh/abelian

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1
Forks
1
Acompanhando
0
Saúde
28%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Triagem de parcelas de terreno para projetos solares em escala comercial
  • Avaliação de risco de wildfire e inundação em portfólios imobiliários
  • Análise de trade-area para expansão retail/QSR com dados demográficos

Veredito

Plataforma robusta e bem documentada que combina rigor metodológico (whitepaper peer-reviewed, validação em 10 estados) com usabilidade prática — MCP server produção-ready para integração em sistemas de decisão geoespacial. Stack moderno (FastAPI + PostGIS + Next.js) com demo aberta, ideal para equipes que precisam de rastreabilidade total em análises espaciais críticas.

README

A tradução para o português deste README ainda está na fila.

Heavi

Deterministic, validated spatial site-screening for renewable-energy development.

Heavi scores candidate land parcels for utility-scale solar suitability against 15 federal and open data sources, and — unlike an AI answer engine — every result is auditable: identical inputs always produce identical outputs, each parcel reports which data source backed each criterion and how confident that source is, and every assessment carries the peer-reviewed methodology behind its weights.

🔗 Live demo: heavi-web.vercel.app — open, no sign-in required.

This is a public portfolio build of a working prototype. The deployed demo runs against a read-only backend; sign-in has been removed so anyone can try the three product modules.


What it does

Heavi is a single spatial platform with three product surfaces:

  • Heavi Energy — solar site screening. Score parcels (one location or a CSV batch) against 14 criteria across 15 federal data sources, see them ranked and color-coded on a map with interconnection-queue context, and export audit-ready PDFs.
  • Heavi Hazard — wildfire and flood risk assessment for property portfolios.
  • Heavi Locations — trade-area analysis (demographics, competitive density, drive-time catchments) for retail/QSR expansion.

How the solar scoring works (in brief)

Suitability is a weighted linear combination of normalized criterion sub-scores, with five binary exclusion screens (critical habitat, protected areas, steep slope, developed land, wetlands). Criterion weights are calibrated per NERC region by constrained optimization against a ground-truth corpus of 6,321 EIA Form 860 operating solar installations. A data selection engine picks the best available source per criterion per location and reports a weakest-link confidence tier — including an explicit "cannot assess" rather than a misleading default when data is missing. Validation across 10 US states shows 71% recall among greenfield-eligible installations with positive discrimination versus matched random land in all 10 states.

Full details: docs/whitepaper/Heavi_Solar_Methodology_Whitepaper_Public.pdf (also served at /whitepaper.pdf on the demo).


Tech stack

Web (packages/web)

  • Next.js 15 (App Router) · React 19 · TypeScript
  • Tailwind CSS v4 · MapLibre GL · lucide-react
  • Clerk (authentication — retained but unguarded in this public demo)
  • jsPDF / html2canvas for client-side exports

API (packages/api)

  • Python · FastAPI · Uvicorn
  • PostgreSQL + PostGIS via asyncpg
  • ReportLab (PDF generation) · SciPy (per-region weight calibration)

Other packages

  • packages/mcp-server — a Model Context Protocol server exposing Heavi's spatial tools (TypeScript)
  • packages/data-catalog — loaders for the federal/open datasets (Python)
  • packages/validation — validation harnesses and methodology generation (Python)

Monorepo managed with pnpm workspaces.

Data sources

15 federal and open datasets feed the solar framework, including NREL PVWatts (solar resource), USGS 3DEP (terrain), HIFLD (transmission), USGS PAD-US (protected areas), USFWS (critical habitat / wetlands), MRLC NLCD (land cover), USDA SSURGO (soils), FEMA NFHL (flood), EIA Form 860 (ground truth), and the LBNL "Queued Up" interconnection queue.


Repository layout

packages/
  web/           Next.js front end (the demo)
  api/           FastAPI scoring + PDF backend
  mcp-server/    MCP server exposing the spatial tools
  data-catalog/  Dataset loaders
  validation/    Validation + methodology harnesses
docs/
  whitepaper/    Methodology whitepaper (public summary)
  specs/         Engineering specs written to drive the build (see below)
  validation/    Validation summaries
  sales/         Sample output package (assessment + portfolio PDFs)

About docs/specs/

docs/specs/ holds the written feature specifications that drove this project. Each was authored as a prompt-style brief and handed to Claude Code, which implemented the feature, validated it, and recorded results under docs/validation/. They're kept in the repo as a record of how the system was scoped and built.


Running locally

The front end builds and runs standalone:

pnpm install
cd packages/web
cp .env.example .env.local   # add your own Clerk + API keys
pnpm dev

The API requires a PostGIS database loaded with the datasets above plus API keys (DATABASE_URL, NREL_API_KEY, …); see .env.example at the repo root. Secrets live only in gitignored .env / .env.local files and are never committed.

This API is rate-limited for the public demo. To run unlimited, clone the repo and run locally.


Built with Claude Code

Heavi was designed and built solo, with Claude Code as the implementation partner — from the scoring engine and data loaders to the validation harnesses, PDF pipeline, and this front end. The specs in docs/specs/ and validation summaries in docs/validation/ document that process.


Heavi is a research/portfolio prototype. The scoring is transparent and reproducible, but it has not been independently audited; assessments are informational and are not a substitute for an engineering interconnection study or professional site diligence.

Ver o original no GitHub

Bom para

Relacionados