Small Project: LLM Learning Assistant

This project implements LLM-powered assistant(s) that can be activated to enhance an education-oriented task or app.

You are expected to implement a working LLM "agent" (no-code options are provided so this requirement may not require actually writing any code) supported by a carefully crafted prompt. However, your user interface for interacting with your agent does not need to be fully implemented, it can again be:

• functional software (using tools such as Flowbite's Svelte Components or React Suite Components)
• interactive wireframes (using tools such as Wordpress, Squarespace, or Wix)
• visual slideshows with written explanations between each (using tools such as Figma, PowerPoint, or ExcaliDraw (a Markdown-oriented drawing tool!))

You may:

• work in groups or on your own
• use ChatGPT or other LLM to help you as long as its contributions are clearly referenced
• extend your APT Update project to incorporate LLM(s)

Submitting your Work

Fork the assignment repository to make your own personal area to work in and use GIT to push your agent to it. Minimally, you must push your progress before both the in-class demo and final submission deadlines.

In addition to the basic information included in the README file:

• explain the subject area and learning goals you are hoping to achieve and how an LLM assistant supports achieving them
• describe in as much detail as possible what kind of learner you are targeting and their needs: novice or advanced students, general or specific subject area, age group (middle, high school, college, or life-long learners), etc.
• describe the context in which the LLM Learning Assistant will operate: as the primary learning option, in support of or embedded in other learning options, in support of self paced learning, etc.
• justify what data about learners using the system you would track by linking it to feature(s) you envision
• justify your UI design, features, and use of an LLM with learning theory or principles
• address ethical, equity, bias, and inclusion considerations in your UI design, features, and use of an LLM
• analyze how your customized prompt(s) changed from their initial to final version
• include impressions or quotes from at least three students (from outside the course) you have asked to interact with your assistant

Record a video, no more than 5 minutes long, going through your assistant. You can use Zoom or there are many free screen capture tools available or free trials of some pretty powerful tools as well.

Specifications

While OpenAI's "chat" interface has proven to be a revolutionary way to introduce many people to LLMs, there are advantages to embedding LLM agent(s) within a broader app that supports more targeted educational goals:

• enhanced security and privacy
• task-specific functionality in multiple contexts
• more flexible user guidance options
• more friendly or consistent UI

Design embedded LLM assistant(s) to improve an existing learning experience or make up your own app to tackle a learning scenario.

You will most likely need to engineering specific prompt(s) to submit to the LLM to tailor its response to your specific task(s). Consider what other text would be appropriate to pass on to the LLM, such as:

• direct input from the learner (if any)
• context from the scenario (part of or the entire current code, writing, or web page.; selected text; etc.)
• fixed text used to fine-tune the LLM (lecture notes, conversation, book, etc.)

You can use any tool you want to build your assistant. I recommend the following:

• MindStudio, free after sign-up, no code required
• OpenAI GPTs, sign-up and $20 fee required, no code required
• LangChain, free open-source, Python coding using GPT APIs
• Fixie, free open-source, web or coding based GPT APIs

If you do not have any specific ideas, feel free to ask your friends or other students.

Here is a presentation about Duolingo Max to give you an example (you are certainly not expected to do something this sophisticated). You are also welcome to look at any other examples online (such as those given in class), but your version should be distinctly different from any you find (i.e., create your own, do not simply copy one).