Lumiere: An AI research assistant

Hello all,

I have been working on an application for a few months now. Being new to Streamlit, it took me some time to figure things out, but I eventually got the hang of it.

The application is called Lumiere, and it’s an AI chatbot designed to make interactions smooth and efficient. Lumiere can handle a alot of tasks, from answering questions and recommending resources to processing file uploads and extracting information. You can upload PDFs, DOCX files, text files, and images, and Lumiere will process and analyze the content for you.

One of the coolest features is the built-in web search and scraping functionality. With a click of a button, Lumiere can fetch up-to-date information from the web and provide you with well-cited, reliable answers. There’s also a chat session where you can continue your discussions with Lumiere and get personalized responses based on your previous interactions.

I would appreciate any feedback as I plan to improve this application. My goal is to make Lumiere more efficient and user-friendly. The best part? It’s free and offers a level of efficiency that rivals many other AI chatbots out there.

Link to the Lumiere App:

1 Like

Wow! Excellent stuff! We are working on almost the same thing independently! Your system is very similar to my Sigma AI (see https://sigmaai.streamlit.app). Sigma AI supports chatting to multiple models simultaneously, but does not yet support the document upload like your Lumiere does – I am planning to add this feature (and image generation too) very soon. I like your choice of expandable “History” with the delete button. In my version the threads are shown as a list and there are separate “Rename” and “Delete thread” buttons.

Btw, in Lumiere in gpt-4o output the formulae are shown mainly correctly via LaTeX, but sometimes incorrectly, see the \hbar shown literally in the main text in the screenshot below:

In Sigma AI I have overcome this problem by adding explicit instruction to the system prompt to use ‘$’ for LaTeX delimiters consistently. I think there is a 1001 little things like this we can share and help improve each other’s system! :slight_smile:

Thanks a lot for your feedback! It’s really cool to see that we’re working on such similar projects. I had a chance to try out your Sigma AI, and I must say, I really like the models you’ve added. But, I did run into a bit of trouble when signing up. Maybe you could consider putting the login and signup on a separate page? That way, users can choose to log in or sign up right when they enter the website, which might make your UI less cluttered. Also, adding a “Logout” button that logs you out and navigates you to the login/signup page could improve things a bit.

Feel free to take any features you find useful from my project. By the way, thanks for pointing out the issue with the LaTeX in Lumiere. I’ll definitely try your suggestion about using the ‘$’ for LaTeX delimiters to fix those formulae issues.

Looking forward to more exchanges! :grin:

You could also try adding custom theming to your app. :slightly_smiling_face:

Yes, thank you for very good suggestions, I will consider them. But what I really liked and am definitely implementing straightaway is your way of presenting threads with the expandable History, only I will have both the “Rename” and “Delete” buttons next to each thread. Also, I will generate the default thread name like in Lumiere and ChatGPT, which would allow the user to go straight to the chat without having to select or create a new thread explicitly. I will study Lumiere more and am sure there will be other things that will be improved in Sigma AI based on it.