Coming from an architecture background, I eventually shifted to IT and never established my own firm.
This time, we are developing an LLM that can provide information on complex building regulations (though it doesn’t have to be limited to building regulations) in collaboration with professionals who regularly review these intricate laws and handle detailed negotiations with authorities.
By limiting its application scope, we believe this system will be user-friendly. If we implant detailed knowledge, focusing on intricate interactions, it could prove to be very useful.
For this project, we simplified the overall system significantly to allow us to concentrate on data input.
The benefits of this approach include:
- Unlike external APIs, there won’t be sudden specification changes, allowing for the use of a fixed LLM version over a long period.
- We can add new learning or perform vector searches to accommodate annual updates to the laws.
- It’s possible to incorporate related regulations.
- Both vector searches and the mathematical models themselves can be fine-tuned for more precise adjustments.