Structural Steel Design using AI

Watch more videos of Claude doing structural steel design using RAG databases

Structural Steel Design using AI 

Since early 2026 the frontier AI LLM models have been capable of undertaking structural engineering calculations. However, care needs to be taken with context - it is well known that providing too much context (eg dumping an entire structural specification into the prompt), leads to reduced performance and even hallucinations. This is particularly true for structural standards, which can be huge documents (eg ANSI/AISC 360 is 800 pages and corresponds to around 1 million tokens of context).

I solved this problem by converting free structural steel specifications and standards from AISC and AISI to RAG databases (DBs). RAG stands for "Retrieval Augmented Generation", and allows the AI to gather all the necessary sections, formulae and tables relevant to the design problem. The AI then writes a python code of these procedures and formulae to calculate the design capacities. RAG is explained in detail on the RAG DBs page.

What is the design workflow when using AI for steel design?
It is up to the engineer to undertake the necessary preliminary design and structural analysis. The list of members and connections and their design loads is collated and passed to the AI (tools to facilitate this information transfer from popular structural analysis software is currently being developed). The AI then accesses the relevant specification RAG and undertakes all calculations required to check the design capacities. Simply ask the AI to prepare a report of the design referencing every specification section and equation, to get a completely verifiable set of design calculations.

To get your AI to do steel design, you will need a RAG DB of the specification(s) you are designing to, an API to connect your RAG DB to your AI, and access to your favourite AI LLM. It is recommended to use the most current frontier model. I use Claude in my videos as I like the easy MCP integration with Claude Desktop.

Step 1: Create the RAG DBs you require for your design from the RAG DBs page.
Step 2: Follow the instructions on the RAG on PC page to set up the API for running queries.
Step 3: Connect your LLM of choice to your RAG DB:
a) Manually - if you use LLMs via their web page, you can manually query the RAG DB via a web UI then paste the result into the query box with your design query.
b) Agentically - connect your LLM to your RAG DB API, then it can perform queries independently. Either: create an MCP wrapper and use Claude Desktop App (see the video examples); and/or create a Python tool for the LLM to access your API and use any frontier model API.

I have provided example code for an API, a web UI and an MCP wrapper on the AI Agent page.