RAG Vector DBs for AI Engineering Agents

Cold-Formed steel structures DBs Structural Steel buildings DBs Stainless Steel buildings DBs

What is a RAG Vector Database (DB) of a specification?
The specification has been converted from a pdf file into chunks of text, equations, tables and figures. The chunks are then embedded into a vector database, which allows for semantic searching. Retrieval Augmented Generation (RAG) is an AI framework that improves LLM accuracy by fetching external, trusted data before generating responses. It bridges domain knowledge gaps, reduces hallucinations, and allows the use of up-to-date, proprietary data without retraining models.

I have the specification already, why do I need a Vector DB of the specification?
You have access to the specification, but your AI Engineering Agent does not. Having your specifications in Vector DB format means your AI Engineering Agent can query the specification and retrieve relevant equations, evaluation procedures and background information (especially when the commentary to the specification is also included in the Vector DB).

When I query an LLM about specifications it usually gets it right already. Why Vector DB?
LLMs do not store specifications. While they can reproduce relevant sections from cross-referencing text books, example calculations etc, LLMs may not have the latest information, they sometimes hallucinate and make errors, and INFERENCE TOKENS ARE EXPENSIVE. With a Vector DB of your specifications on your personal device, you and your AI Engineering Agent can make infinite queries without using any tokens, then feed the expensive LLM with the relevant information directly for the LLM to act on.

What's the workflow when using a Vector DB specification?
The general workflow is that when the user (human or AI Agent) makes a query, the query is first routed to your RAG system to fetch trusted information directly from the specification. The usual use cases for engineering specifications apply: engineering design calculations, checking designs, learning about design principles and calculations, education and training about new specifications for new users and/or new editions. The query and the relevant information from the Vector DB retrieval are then fed as the prompt to your expensive LLM.

What's the difference between me searching or my AI Agent searching?
You can create a simple web UI to undertake manual searches, put the query and the RAG retrievals into a text file, then manually upload to your favourite LLM. Or, you can generate an API with a simple python code, point your AI Agent to the API, and it can undertake queries independently.

How to setup a RAG Vector DB?
Download Vector DBs below for particular specifications and locate on your personal computers, shared networks and/or cloud servers. The instructions for how to setup the RAG query API are on the Engineering Rag page.

What resources are required?
Computing resources are very minimal and the system can be setup on Windows, Linux or Mac. The RAG system uses Qdrant, Docker, nomic-embed-text and FastAPI, which are all open source/free apps. The zipped Vector DB files are all less than 10 MB each, and a retrieval of a RAG query is virtually instantaneous and requires very little RAM. The entire setup is free.

Note on Specification images: In the interests of keeping the file sizes small, images have not been retained in the Vector DBs. You should get the free pdf downloads of the Standards from AISI and AISC so you can reference figures if required. Only Specifications that are distributed for free by these industry bodies have been included in these RAG DBs. The free RAG DBs are provided here to promote the use of steel in buildings.

RAG Vector DBs for Cold-Formed Steel Structures

North American Specification for the Design of Cold-Formed Steel Structural Members: AISI-S100-22 RAG file

Standard for Cold-Formed Steel Framing - Prescriptive method for One- and Two-Family Dwellings: AISI S230-19 RAG file

Code of Standard Practice for Cold-Formed Steel Structural Framing: AISI S202-20 RAG file

North American Standard for Cold-Formed Steel Structural Framing: AISI S240-20 RAG file

North American Standard for Thermal Transmittance of Building Envelopes with Cold-Formed Steel Framing: AISI S250-22 RAG file

North American Standard for Seismic Design of Cold-Formed Steel Structural Systems: AISI S400-20 RAG file

North American Standard for Cold-Formed Steel Nonstructural Framing: AISI S220-20 RAG file

RAG Vector DBs for Steel Structures

Specification for Structural Steel Buildings: ANSI/AISC 360-22 RAG file

Seismic Provisions for Structural Steel Buildings: ANSI/AISC 341-22 RAG file

Prequalified Connections for Special and Intermediate Steel Moment Frames for Seismic Applications: ANSI/AISC 358-22 RAG file

Seismic Provisions for Evaluation and Retrofit of Existing Structural Steel Buildings: ANSI/AISC 342-22 RAG file

Code of Standard Practice for Steel Buildings and Bridges: AISI/AISC 303-22 RAG file

RAG Vector DBs for Stainless Steel Structures

Specification for Structural Stainless Steel Buildings: AISI/AISC 370-21 RAG file