LLM-Enabled Antenna Modeling

Large Language Model-Enabled Antenna Modeling

LEAM is an open-source research toolkit for parameterized antenna modeling. It supports both CST and HFSS, provides a Windows desktop interface for interactive use, and can connect naturally with downstream simulation and optimization workflows.

LEAM works from textual and image-based antenna descriptions drawn from technical reports, research papers, patents, and other engineering documents, helping users move from descriptive source material to analyzable EM models.

LEAM can export parameterized CST and HFSS antenna models together with structured intermediate representations such as parameters and geometry-related modeling data. This makes antenna models easier to store, review, reuse, and extend, while also supporting the digitization of legacy antenna design data and the activation of underused engineering assets.

Overview

A practical stack for LLM-assisted antenna modeling

LEAM combines an LLM layer, simulator-facing modeling workflows, and structured intermediate data so that antenna descriptions can be turned into reusable EM modeling assets rather than one-off generation results.

In practical terms, the stack is built around OpenAI-driven multimodal understanding, CST and HFSS support, a Windows desktop interface, and structured parameter or geometry data that can be stored, reviewed, and reused across projects.

LLM Layer

LEAM currently uses OpenAI models for multimodal antenna understanding and model generation, with room for future LLM API support.

Simulator Layer

LEAM supports both CST and HFSS, translating antenna descriptions into EM models inside the tools engineers already use.

Workflow And Data

A Windows desktop GUI and structured parameter or geometry data make modeling sessions easier to review, store, and reuse.

Navigation

Docs, desktop guidance, and API entry points

The README gives the short version. The docs set is split by use case so antenna engineers can go straight to desktop onboarding, while script users can jump directly to the Python surface.

The main documentation now lives in three focused pages: desktop onboarding, workflow reference, and Python API.

Desktop

Install LEAM, set OPENAI_API_KEY, launch leam-desktop, and understand the normal first-run path.

Open getting started

Workflow

Read the exact desktop behavior for workspaces, optional branches, attachment handling, and execution gates.

Open workflow reference

API

Use the Python API page for backend wrappers, runtime helpers, and programmatic access to built-in examples.

Open Python API

Workflow

A workflow layer between description and EM simulation

LEAM organizes antenna modeling into staged generation, review, and optional execution steps.

Designer Initial
Solids
Parameters
Materials
Solids
Decision Check
Solid
Dimensions
Optional Run
Update
Optional Parameter
Update
Run Simulator
Boolean
Optional 2.5D
Model
3D Model

Execution stays separate so intermediate results can be checked and revised before simulator runs.

Designer Corresponding to the `weak_description` workflow in the GUI. Optional Enabled only when the selected workspace settings turn that branch on. Decision Review step that can route issues back upstream before model generation continues.

Quick Start

Launch LEAM from Python 3.11 on Windows

LEAM Desktop currently targets Windows, requires an OpenAI API key, and needs at least one local CST or HFSS install.

Normal users do not need a virtual environment or a separate configure step.

LEAM / desktop quick start
$ pip install leam
[ok] Installed LEAM Desktop with CST/HFSS support in the base package
$ $env:OPENAI_API_KEY = "your_api_key_here"
$ leam-desktop

LEAM will auto-detect CST and HFSS on first launch. If `leam-desktop` is not on PATH yet, run `py -3.11 -m leam.desktop`.

LEAM desktop GUI preview