CountryPlans.com
 

Medeek Project Plugin

Started by Medeek, January 14, 2023, 04:50:14 PM

Previous topic - Next topic

0 Members and 2 Guests are viewing this topic.

Medeek Engineering

Interior Bearing (ie. footing + post & beam) are now added to the House Builder.



Nathaniel P. Wilkerson

Medeek Engineering

I've also enabled the entry stair option for foundation door openings.

Nathaniel P. Wilkerson

Medeek Engineering

I'm also adding gable walls as well as shed and hip.

Nathaniel P. Wilkerson

Medeek Engineering

Currently the various plan view tabs are treated as their own "sheets" or pages. In other words if you print that tab you get exactly what is in the viewport scaled to fit into a regular sheet size (letter, 11x17, Arch C, Arch D). You cannot add in other geometry from other tabs. This seems to work reasonably well for most of these plan views (ie. Floor Plan, Foundation Plan, Roof Plan, Floor Framing Plan etc...)

However, when it comes to elevation views things typically get a bit more interesting. Some designers like to put multiple elevation views on one sheet and in some cases all four if they will fit ( I like to do this for small ADUs for example). So it is not as easy as assigning one elevation view to one sheet, or even assigning two elevation views to one sheet, there needs to be more flexibility.

After giving this some considerable thought I decided what I really needed was a Layout system where the user can drag the elevations into the "workspace" and position them as they see fit, and then at that point add their annotations and dimensions. I have now created four additional Layout tabs where the user can do exactly that.

The only major thing I feel like I am missing at this point is a sectional view or section tool.





Nathaniel P. Wilkerson

Medeek Engineering

I've also add Common Truss Roofs along with Complex Roofs.





Nathaniel P. Wilkerson


Medeek Engineering

**Version 1.2.8** - 10.11.2026
- Added the "Draw House" tool to the Medeek Builder toolbar.

I'm releasing this update so that interested parties can test out the House Builder interface.  It is currently not fully wired into the Medeek API so have temporarily disabled the "Create House" button.  I will be spending the next week or two completely the upgrades to the Medeek API for the various mdkBIM extensions so that we can access API methods for floors, foundations, roofs etc...

The primary purpose of this update is to garner feedback on the UI and the front end.  Please download and install this latest update and send any feedback directly to nathan@medeek.com.  I will be very curious what everyone thinks of this new tool and I think you will find it is much more capable than the "prototype" Garage Builder".
Nathaniel P. Wilkerson

Medeek Engineering

Version 1.3.0 - 10.19.2026

Wired up the Medeek API with the House Builder so that assemblies can be created with the following extensions: Wall, Truss, Foundation and Floor.

Note that in order for the House Builder to work correctly you will need to upgrade your Wall, Truss, Foundation and Floor extensions to the latest versions (also released on 10.19.2026).



Nathaniel P. Wilkerson

Medeek Engineering

**Tutorial 8** - House Builder (19.42 min.)

Nathaniel P. Wilkerson

Medeek Engineering

The House Builder is now building all wall-hosted electrical fixtures:



Nathaniel P. Wilkerson

Medeek Engineering

Tutorial 9 - Electrical Plans (14.41 min.)



Unfortunately I forgot to show the electrical legend in this tutorial.
Nathaniel P. Wilkerson

Medeek Engineering

First look at the Block Editor and Block Library Manager.




Nathaniel P. Wilkerson

Medeek Engineering

Nathaniel P. Wilkerson

Medeek Engineering

I've spent about three days putting together an experimental AI harness for the House Builder it is actually quite interesting to see it run.  Even more interesting than actually running this AI tool is actually having spent the time to develop it. 

Ultimately I had to us AI to build the AI harness and the reason behind this is that I don't actually know how an AI reasons or thinks about a problem, or what information it wants or needs.  So the best way to design this harness was to ask ChatGPT (GPT 6 Astra) what tools or information would an LLM best understand and work with.  After some back and forth and and a few false starts we arrived at a minimal and efficient toolset with which the AI/LLM could manipulate the model inside of the House Builder (Floor Plan only for now).  I would call this development project a collaboration, however most of the edge cases were detected by me and then rectified by the AI.  It honestly feels like I have an employee now, I will say it does seem a bit strange.

For something like this to work you have to build a "spatial model" that the AI can understand and fully comprehend, a clear and concise picture of what it is dealing with.  In other words it needs to be able to manipulate the model and then instantly get feedback (especially if there are errors or it tries to do something illegal).  Also when it does do something illegal you need to explain why it was illegal in very definite terms, otherwise it cannot recover.

Bottom line is that the harness is paramount to providing a successful connection between the AI engine and the House Builder interface.  If this is not created carefully and provides logical and concise data, the AI will be running blind and will makes all kinds of unforced errors.  This was even more apparent when I tested weaker models like Luna and Haiku.

To be able to run this tool you will need to have an API key with at least one provider (OpenAI or Anthropic).  Personally I prefer OpenAI's API implementation over Anthropic, but they are close in performance.  I also think OpenAI's cost per token is considerably cheaper than Anthropic's right now, so that is something worth taking into consideration.

My honest assessment at this point:  I don't think AI has the experience and reasoning to do really effective architectural space planning.  It can follow directions well enough and handle simple structures but I don't think it is yet capable of designing your next "dream home".  So for now, we are still safe, but in a year or two I may have to revisit this assessment.



Nathaniel P. Wilkerson

Medeek Engineering

I've got mdkAI working fairly well.  However my own interface can never be as good as the regular chatbot interfaces so I decided to add mdkMCP.  Currently I'm testing the MCP integration with Claude.  After a bit of coercion I got it to faithfully duplicate a s simple floor plan quite accurately.



Nathaniel P. Wilkerson

Medeek Engineering

Nathaniel P. Wilkerson

Medeek Engineering

As cool as this AI thing is I still think it is easier to just manually "drive" either the House Builder or any of the plugins. I'm not saying that things won't change. If AI keeps improving there is a good possibility that eventually these AI enabled tools may find real utility.

I don't think we are quite there yet but I see the early signs of this when in the tutorial video the AI recognized that the window was no longer properly centered on the wall and offered to fix it for me.

At some point in the future I could see the AI doing a thorough audit of every design, looking for physical clashes or even checking the engineering of every beam, header, floor joist, rafter etc... and then flagging members that look suspect so that the engineer or designer can focus on potential trouble spots.

What you are seeing with my first foray into AI automation is only just the beginning. This is just a proof of concept. It shows that we can create a viable harness which allows the AI to visualize the model, reason about it and then provide it a carefully curated set of tools with which it can then manipulate and mutate the model as it sees fit.

What I did not demonstrate in this video is the DXF "reading" ability I have given my mdkAI. I don't know that the regular AI chatbots have this sort of ability yet.

What I first considered was building a DXF renderer which would allow the AI to "visualize" the DXF (much like passing the AI an image or a rasterized PDF), however this is not entirely deterministic and prone to error. After giving this some additional thought and bouncing my ideas off of ChatGPT 5.6 Sol we both decided that a better solution was to actually import the DXF and then decompose it into a well organized "database", that the AI could then query on demand. This solution turns out to be very fast and very accurate (no guessing). I may need to do a video just showing how this actually works, it is really quite impressive.

Right now it feels like the late 90's during the DotCom era. We are still feeling out this new technology and trying to figure out where it works and where it doesn't. It is still very much early days.
Nathaniel P. Wilkerson

Have a question about your own build?

Sign up free and ask the members. It takes a minute, and we email you a link to confirm.

Sign up freeAlready a member? Log in or reset your password.