Vector: A Docling + RAG Workflow Researching Documents (w/ Gradio UI)

[dkylewillis]

Hey everyone, my name’s Kyle. I’m a civil engineer out of Georgia, but I’ve always had a side passion for making my work more efficient through automation. Python has been my main tool for years, and lately I’ve been diving into AI to see how it can help.

One of the biggest pain points I run into is the due diligence phase—digging through ordinances and technical manuals trying to find the one line that can hold up a whole project. Miss it, and you can waste a lot of time and money.

To tackle that, I built a project I call Vector. It’s a RAG workflow that uses Docling to convert documents into structured text, embeds them with Sentence Transformers, and stores everything in Qdrant. Figures and tables get saved as PNGs so they show up with the answers too.

I also put together a CLI and a Gradio web app. It works kind of like Google’s NotebookLM, but I’ve had much better results with the vector approach—especially since it can bring back the actual figures along with the text. That’s been a huge help when searching technical documents.

[kevinschaich]

this is cool @dkylewillis - do you intend for it to be open-source/public? The link in your post is broken – you may have the repository settings on private (not sure if that is intentional).

[dkylewillis]

I just made this repo public: https://github.com/dkylewillis/vector

It’s probably a bit early to share since the README still needs work and there are several features and improvements I’d like to add. But I only have about 8–10 hours a week to spend on it.

[az-ihsan]

hi @dkylewillis, this project is rad! I have been looking for a similar problem to index pdfs of manuals and safety regulations in my domain. I think your project suits me here. I will take a look into your repo and maybe in the future will add the contribution :)

[dkylewillis]

Thanks! I'm hoping to improve it a lot over next couple months. Any help i can get would be grealty appreciated.