Speckle LLM Agents: Inspiration for Speckleverse

Hi guys,

I’ve been working on a multiple LLM agent setups for Speckle data. Some parts are easier to handle with PandasAI, while others are more sophisticated.

Unfortunately, I don’t have time for Speckleverse this weekend. Super sad!

It is definitely not perfect yet, but I thought this could be helpful for some of you. So I decided to publish it and share some open-source power! :wink:

You can find under: GitHub - jjanixx/bim_ir_chat: LLM agents for BIM Information Retrieval

If anybody needs a starting point for their project, maybe this will help. I tried to write a comprehensive README, but if something is unclear or if you encounter any bugs while copying the code, please let me know.

If there is interest, I can create a Loom video describing the prototype and the code.

Have a great hack! :rocket:


Hi @janix, super thanks for the shout and reference - i’m betting it’s going to be useful. My curiosity has been activated :slight_smile:


Hey @janix!

Looks like a cool starting point. I will look into your repo and see if I can collaborate with you on this if it’s okay with you. Maybe not for the hackathon, but going forward for sure.

I am also learning Langchain Agents and want to integrate tools to work with BIM data.



Nice, looking forward to your suggestions and discussions! :grinning:

1 Like

Rereading I post; I saw I did not fully describe why it could help you guys:

  • I have pretty simple off-the-shelf LLM agent by PandasAI (in the end 3-5 lines of code) → works pretty well for simple use cases (such as questions, diagrams, table adjustments about certain Speckle categories)
  • if you want to create something more complicated, I started on developing Langchain agents with different tools appended (this super flexible and you can tweek and debug endlessly, probably performance is right now degraded compared to the PandasAI chatbot, but in the end it should be way more powerful, for this one you need a bit more knowledge about LLM agents)

Have fun! Super happy if it helps anybody! :grinning:


Nice impressive job thanks for posting. I have a similar idea for querying models for the hack but want to use a semantic-kernel since I’m more C# than Py :smiley:


Hi @janix,
FYI: I created a fork of your repo bim_ir_chat and have started experimenting with it by creating notebooks in a separate folder called “research”.

To begin with, I re-created the RAG history aware Q/A chatbot. I added the Canadian National Building code for context.
Here it is → bim_ir_chat/research/langchain-test-01.ipynb at 48e393e0e29146a143cb16cf045b894a4623753e · bhargobdeka/bim_ir_chat · GitHub

I will keep you informed!:tipping_hand_man:t4:



Nice, looking forward to see what you will come up with!


Impressive work ! :clap: