Hivemind

landing page
sample lesson
workflow:)

Inspiration

The inspiration for Hivemind stemmed from personal frustration with the quality of available lectures and resources, which were often insufficient for effective learning. This led us to rely entirely on ChatGPT to teach ourselves course material from start to finish. We realized the immense value of tailored responses and the structured learning that emerged from the AI interactions. Recognizing the potential, this inspired the creation of a platform that could harness collective student input to create smarter, more effective lessons for everyone.

What it does

Hivemind is an AI-powered learning platform designed to empower students to actively engage with their course material and create personalized, interactive lessons. By allowing students to input course data such as lecture slides, notes, and assignments, Hivemind helps them optimize their learning process through dynamic, evolving lessons. As students interact with the platform, their feedback and usage patterns inform the system, organically improving and refining the content for everyone. This collaborative approach transforms passive learning into an active, community-driven experience, creating smarter lessons that evolve based on the collective intelligence and needs of all users.

How we built it

Backend: Developed with Django and Django REST Framework to manage data processing and API requests.
Data Integration: Used PyMuPDF for text extraction and integrated course materials into a cohesive database.
Contextual Search: Implemented Chroma for similarity searches to enhance lesson relevance and context.
LLM Utilization: Leveraged Cerebras and TuneAI to transform course content into structured lessons that evolve with user input.
Frontend: Created a React-based interface for students to access lessons and contribute feedback.
Adaptive Learning: Built a system that updates lessons dynamically based on collective interactions, guiding them towards an optimal state.

Challenges we ran into

Getting RAG to work with Tune
Creating meaningful inferences with the large volume of data
Integrating varied course materials into a unified, structured format that the LLM could effectively utilize
Ensuring that lessons evolve towards an optimal state based on diverse student interactions and inputs
Sleep deprivation

Accomplishments that we're proud of

Functional Demo
Integration of advanced technologies
Team effort

What we learned

Throughout the development of Hivemind, we gained valuable insights into various advanced topics, including large language models (LLMs), retrieval-augmented generation (RAGs), AI inference, and fine-tuning techniques. We also deepened our understanding of:

Tools such as Tune and Cerebras
Prompt Engineering
Scalable System Design

What's next for Hivemind

Easy integration with all LMS for an instant integration with any courses
Support different types of courses (sciences, liberal arts, languages, etc.)
Train on more relevant data such as research studies and increase skill level of the model
Create an algorithm that can generate a large amount of lessons and consolidate them into one optimal lesson
Implement a peer review system where students can suggest improvements to the lessons, vote on the best modifications, and discuss different approaches, fostering a collaborative learning environment