bluegreencity-ki

Infrastructure Overview

Last updated: 2026-02-06

This document summarizes the technical environment used to run BlueGreenCity-KI on a dedicated workstation. It is intended to explain the architecture and make it reproducible without exposing live infrastructure identifiers.

Hardware Profile

The project runs on a workstation-class machine with:

recent multi-core CPU
high-memory configuration suitable for large raster processing
professional GPU for local model experimentation
fast SSD storage for source data, derived rasters, and logs
external backup storage for periodic snapshots

Core Software Stack

Area	Main Components
Database	PostgreSQL 17, PostGIS 3.5, Docker Desktop
Python	Conda-based Python 3.11 environment with GeoPandas, Rasterio, NumPy, pandas, scikit-learn
GIS	QGIS 3.40+ and GDAL
Development	VS Code with Remote SSH and GitHub integration
Local AI Tooling	LM Studio and related local inference tools

Network and Access Pattern

The workstation is accessed through a private networking layer. The recommended pattern is:

VPN-based private connectivity
SSH for development access
optional Remote Desktop for administrative or GUI-heavy tasks
GitHub for source control
managed file sync only for non-sensitive working data where appropriate

Data and Service Layers

The system is organized into four operational layers:

source data storage for large raster and vector inputs
processing workspace for derived intermediates and logs
PostGIS database for structured project data and analysis results
client tools such as DBeaver, QGIS, and Python scripts

Logical Database Areas

Prefix or Schema	Role
trees	tree inventory and species-related attributes
context	supporting context layers such as streets and land use
analysis	risk scores, pipeline outputs, and reporting tables
climate	imported climate and environmental layers

Reference Architecture

Private network or VPN
    |
    +-- Remote development client
    |      +-- VS Code Remote SSH
    |      +-- DBeaver
    |      +-- QGIS
    |
    +-- Workstation
           +-- Docker Desktop
           |      +-- PostgreSQL/PostGIS
           |      +-- Optional pgAdmin
           +-- Conda Python environment
           +-- Local source and derived data storage
           +-- Optional local LLM tooling

Operational Principles

Keep credentials out of repository files and public documentation.
Prefer stable private hostnames over hard-coded IP addresses.
Treat the database as the shared system of record.
Keep large inputs and derived rasters outside Git.
Use scripted workflows for repeatable imports, scoring runs, and exports.