Actually start experimenting and implementing the AI Citation Helper UI.

I need to decide on a high level description/plan for how I'm going to design the AI Citation Helper UI.

This will mean a lot of explaining, thinking about what requirements I need, weighing different options, and designing at a high level.

1. Describe more deeply what the AI Citation Helper is
2. Explain some of the requirements
3. Explain what work I've already done
4. Start planning furtther implementation of features.

Okay so I need to explain more about the AI Citation Helper. It's a standalone LLM tool I've been building for Contoural. Contoural works with clients to help them understand what legal and regulatory requirements their clients face, and also help them develop plans and processes to be legally compliant with them. One of the most fundamental services Contoural offers is citation work. Citation work is where Contoural will help a client compile a list of citations to legal text, which contain exactly all of the legal and regulatory rules on how their specific data should be saved. It can be a very difficult and time consuming process to determine which citations are relevant to a given client. The AI Citation Helper is a tool used by Contoural to reduce the time and effort needed to determine which citations are relevant to a client's business.

The AI Citation Helper tools is a program that takes in raw citation data from a third party database, as well as settings for a specific client company, and then uses LLMs to classify and clean the raw data. It returns a filtered excel sheet which pre-classifies each citation as being "relevant", "not_relevant", or "further_research_needed". This is important, as the process of verifying "relevant" and "not_relevant" is incredibly fast. The user can focus on the citations which need research, and utilize further LLM tools to assist them in this. In initial testing, this system saves 50% of time. However, currently there is no frontend for a Contoural employee for using it. Right now, I'm manually uploading provided excel sheets and then pushing the button to run the system myself. I need to develop some self-contained application so that I can remove myself from the process entirely.

So there's a couple of requirements for the system which are important. The number one requirement is that it has to be easy to use. I'm going to be sending this to domain experts who aren't as comfortable with highly technical work as I am. However, they are very comfortable with excel and doing massive amounts of work in it. This leads to the next requirement, which is I would like the system to take in excel as input and output excel. This will make it integrate easily with Contoural's existing processes. Another requirement for the system is simplicity. I don't want to simply clone excel and add AI features, I want a more targeted and streamlined application. In this case, there may be very few cases where I actually need EDITING rows to be allowed.

So let's talk about what I've currently built. I've taken some inspiration from my daily blog builder tool and decided to build the frontend using InstantDB and NextJS, while also using a Flask Python backend for AI features. Right now it's in an incredibly simplistic state. It shows a fake "excel" table for a single data set, with some not working features for switching "sheets" by classification.

I want to add functionality to make viewing the different sheets smoother. I'd like to add advanced filtering and sorting capability to the columns. I need to add some fast and easy way of verifying AI output. Speed is key here, Contoural employees are going to be verifying 1000s of rows of AI classifications, as well as using the tool to conduct their own research.