The below blog is put together by Sourav Singh from Quest Alliance (sourav.singh.dev@gmail.com)

What happened?

The LLM cohort at Kochi included :

• NGO showcase ~ showcasing their work with LLMs and/or what they want to do.
• Expert Talk ~ Talk from experts on working on LLMs & prompt engineering.
• Work hours ~ Open hours to collaborate/prototype.

What we did?

Besides discussion and cross learning, we were able to hack a flow to experiment with personalized learning journeys where …

• user shares details ( location, age, hobbies, what he/she wants to learn )
• based on the input, an explanation is generated.
• Then, a coding question is generated & the user is asked to submit a solution.
• The submitted solution is analyzed, and feedback is shared on it.

Click here to try it out. Refer to attached screenshots ( last page ) to see examples.

Takeaways?

Don’t let technological considerations restrict experimentation

The conversations sparked by Edmund’s presentation were pivotal because they provided valuable insights into the world of Large Language Models (LLMs) and shed light on the approach we should adopt when embarking on experimentation or solution-building. It seemed that many of us, particularly those involved with NGOs, were constraining our ability to harness the potential of LLMs by imposing self-imposed barriers such as concerns about costs, self-hosting, or the search for alternatives. This served as a valuable reminder to us all to release our preconceptions and not allow technological considerations to restrict our creativity in designing solutions.

Prompt engineering should be prioritized .

Another intriguing discussion revolved around my conversation with Aman from HyperVerge. Their use case closely paralleled ours, and they have already developed a platform that harnesses Large Language Models (LLMs) to generate assessment questions based on provided textual or video content. Aman’s presentation, sharing his insights gained during the platform’s development, proved to be highly informative.

In today’s landscape, organizations often desire AI-enabled solutions that simply deliver results without the need for extensive iteration and prompt engineering tailored to their specific use cases. Aman effectively addressed this challenge, and we even managed to create a simplified flow where we refined custom prompts through several iterations to move closer to the kind of personalized journey we want learners to experience.

This experience served as a stark reminder that LLMs are not a one-size-fits-all solution and highlighted the considerable time and effort required for customization in specific use cases. Witnessing the potential of Aman’s platform was genuinely motivating, showcasing what can be accomplished with dedication and effort. Undoubtedly, prompt engineering stands out as a crucial aspect that all of us should prioritize.

What Next?

Engaging in discussions about LLMs with experts and fellow NGOs was indeed thought-provoking, particularly in the context of our LLM utilization. Our primary objective is to employ LLMs for tailoring personalized educational content to students. 

Additionally, we are exploring the potential of using a chatbot powered by LLMs to address inquiries related to Quest Alliance in a comprehensive manner. Previously, we relied on static content and videos for this purpose, but transitioning to an AI-driven chatbot represents the logical next progression in our approach.

Thoughts on the sprint

The sprint in Kochi was excellent. Tech4dev has a unique approach to sprints where it goes beyond mere work – you engage in conversations, interact with others, and, almost unknowingly, make significant strides in your tasks.

About Us

Quest Alliance is a nonprofit organization dedicated to empowering young individuals with essential 21st-century skills through the facilitation of self-directed learning. We’ve developed a chatbot designed to deliver bite-sized, self-paced educational content focused on teaching algorithmic thinking.

Currently, we’re exploring ways to enhance the learning journey by tailoring it to the unique needs of each learner and are actively engaged in experimentation in this regard.

You can check this in-depth document, for a detailed description of our use case.

Screenshots

Here are fwe screenshots from the chatbot flow we designed as part of the LLM sprint in Kochi. The objective was to explore methods for tailoring the learner’s experience. Initially, we provided users with the option to request content rephrasing, analogies, or examples. However, as we delved into prompt analysis and experimentation, we decided to shift to a fully LLM-driven approach to gain insights into its potential usability.

P.S It is powered by jugalbandi.