BraeinAi Geospatial is a fine-tuned version of Meta’s LLaMA-2 model for Geographic Information Systems (GIS) tasks.
It is designed to assist with spatial data concepts, GIS software usage, standards (OGC WMS/WFS/WCS), and related technical queries.

Developed by: Ispatialtechnosolutions
Model type: Causal Language Model
Language(s): English
License: Apache 2.0
Base Model: LLaMA-2 (Meta)
Fine-tune method: LoRA → merged weights
Intended use: GIS assistant (ArcGIS/ArcPy/QGIS/WMS/WFS/Portal)

Model Details

Model Description

This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.

Developed by: Ispatialtechnosolutions
Funded by [optional]: [More Information Needed]
Shared by [optional]: [More Information Needed]
Model type: Causal Language Model
Language(s) (NLP): English
License: Apache 2.0

Model Sources [optional]

Repository: Ispatialtechnosolutions/BraeinAi-Geospatial

Uses

Direct Use

Answering GIS-related technical queries
Helping with geospatial software usage (e.g., QGIS, ArcGIS, GDAL)
Assisting with OGC standards (WMS, WFS, WMTS, etc.)
Supporting spatial data processing and remote sensing tasks

Downstream Use [optional]

Integrating into GIS chatbots
Embedding in decision-support tools for spatial analysis
Educational use in GIS learning environments

Out-of-Scope Use

Integrating into GIS chatbots
Embedding in decision-support tools for spatial analysis
Educational use in GIS learning environments

Bias, Risks, and Limitations

May generate hallucinated commands for GIS software not grounded in documentation
Limited to English-language queries
Not a replacement for domain experts in mission-critical applications

Recommendations

Validate outputs before production use
Use in supervised / decision-support settings, not as final authority

How to Get Started with the Model

python

from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

repo_id = "Ispatialtechnosolutions/BraeinAi-Geospatial"

tok = AutoTokenizer.from_pretrained(repo_id)
model = AutoModelForCausalLM.from_pretrained(repo_id)

pipe = pipeline("text-generation", model=model, tokenizer=tok)

print(pipe("How do I publish a WMS in GeoServer?", max_new_tokens=200)[0]["generated_text"])

## Training Details

### Training Data

This model was fine-tuned on a curated GIS domain dataset including:

Spatial standards (OGC WMS/WFS/WMTS/WCS docs)

GIS tutorials & training manuals

QGIS/ArcGIS user documentation

Geospatial developer blogs

### Training Procedure

Base Model: LLaMA-2

Method: LoRA fine-tuning

Precision: bf16 mixed precision

Batch size per device: 1–2

Gradient accumulation steps: 8–16

Effective batch size: 4

Epochs: 2
LR         = 2e-4

## Evaluation

Metrics

Perplexity: Lower than base LLaMA-2 on GIS test set

Qualitative evaluation: Produces domain-specific and contextually relevant answers

Example Query & Response

Input: "What is the difference between WMS and WFS?"
Output: "WMS (Web Map Service) delivers rendered images of maps, while WFS (Web Feature Service) delivers vector features in formats like GML/GeoJSON for analysis."

## Environmental Impact

Training hardware: A100 128GB RAM GPUs
24gb Graphic Card

## Technical Specifications [optional]

### Model Architecture and Objective

Architecture: LLaMA-2 (Causal Decoder-only Transformer)

Parameter count: Same as LLaMA-2 base used

Library:  Transformers

## Citation [optional]

@misc{llama2gis2024,
  title = {BraeinAi-Geospatial: A Domain-Specialized GIS Language Model},
  author = {Ispatialtechnosolutions},
  year = {2024},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/Ispatialtechnosolutions/BraeinAi-Geospatial}},
}

## Model Card Contact
Maintainer: Ispatialtechnosolutions
Email: connectus@ispatialtec.com