My boss is asking for me to dig into this a bit. We are a commercial general contractor and specialize in industrial tilt up warehouses.
We do a lot of highly conceptual estimating for our clients since our product price point is pretty consistent in terms of $/sf (with gradual market fluctuations.. outside of 2020-2023), it's mostly sitework that can vary a bit, but that's easy to separate from the vertical portion.

I'm wondering what capabilities chatgpt would have for us to search internal data so when boss gets the phone call from developer and says 'hey looking at a piece of land, how much would 2 buildings about 150ksf each and a single 250ksf be?' For now, he calls me, and if I'm available, I have my own excel program I created that uses unit prices which can be applied to quantities of scope based on the dimensions of the building. For example, a 200x700 building 32' clear height, I can get a lot of info, like foundation perimeter, tilt panel sf, lengths of gas and water piping, qty of code min. unit heaters in warehouse, approx sf of storefront, approx numbers of man doors and OH doors at the dock, etc.. Many items are strictly in a $/sf, like steel, slab, roofing, sprinkler, etc, so it's easy to get a high level $/sf with only those 3 inputs.

But boss has tasked me with either showing him that my solution is easier and just as accurate vs something AI can do, and can be carried on in either of our absences.

Any ideas or which direction to go to get info?? I know absolutely nothing about chatgpt.

Hey so what details can you share on the data you would be searching...

where is it stored? single repository?
based on your use case I assume alot of this data is in excel spreadsheets? is there a standard template for all of this data?

You'll have to dig in and learn.

Your goal is prob a natural language interface to query and get answers.

You won't get that from any out of the box UI. Depending on how big the data set is, how often the data is updated, and if it's accessible via a API or if it's proprietary, you'd go in different directions. Assuming it's probably >100k tokens of data, best place to start is learning about vector databases and setting up a RAG pipeline for a model. Basically this stores your data in a way that can be served to your LLM for inference in a token efficient manner.

Long story short, yes you can absolutely achieve that goal and don't need an engineer to do it, but there is no out of the box solution.

AI won't be any more accurate on its own than your Excel model. What it sounds like you're looking for is an AI agent that can run the same calcs as your spreadsheet. That's doable but probably not cost effective unless you're spending the majority of your time in your spreadsheet building estimates.

fulshearAg96 said:

---

Hey so what details can you share on the data you would be searching...

where is it stored? single repository?
based on your use case I assume alot of this data is in excel spreadsheets? is there a standard template for all of this data?

---

yes we do all of our estimates in excel.
All data is stored on an in house server. We are small outfit.. nothing fancy.

In my mind, using AI to accomplish seems a bit over the top and complicated, considering I've done it with a basic excel spreadsheet with a few user inputs, but boss is asking to at least show him why I say this. I just know nothing about chat gpt and have never used it...

Are you wanting to find an AI solution to this or hoping there isn't one?

It's been said above but I don't believe ChatGPT specifically is the answer to performing that, but I've only used the unpaid version.

Agreed that a free version won't have the capacity to provide what the OP is trying to achieve, or not very easily. I believe the free version is up to maybe 2023, and it seems it limits compute output. I have not seen the Pro version at $200/month but the $20/month version is definitely an upgrade from free. I think with the right prompting, ChatGPT can produce something similar if not identical to his worksheet. It'll even do it in Excel. It's both amazing and sobering what it can produce, so I imagine the Pro version is akin to being on steroids. Would love to have someone comment if they have experience with Pro.

What you want is a RAG solution that is an llm targeting your specific data. What is the data stored in today?

Look at google notebook lm. You can ground it on your specific data then interact with it.

If the data is in SharePoint m365 copilot has embedded agents you could customize and point to specific SharePoint sites or one drives.

We use a tool called glean that connects to our data and I believe would help with your use cases. Not sure the cost or config effort though.

We do technology consulting so it helps with proposals and the like.

If you can give me an email address, I can have this exact conversation with you with some relevant industry insight.