Knowde

Inspiration

The availability of information on the internet is overwhelming. Each member in our group has an insatiable curiosity and desire to learn but sometimes struggle on getting started on a specific topic. We wanted to create an AI Agent that can automate generating a structured learning plan, track progress and understanding, and adapt to your learning style based on your study activity patterns.

What it does

Knowde can generate comprehensive learning plans from vague prompts. Simply tell Knowde that you want to learn about a certain topic and it will provide a goal-oriented, modular lesson plan with reminders to keep you on track on your journey.

How we built it

The backend was developed with Agno to orchestrate collaboration between different agents, FastAPI, and Firestore to maintain a User Knowledge graph and different lesson plans. The frontend was built using react, next.js, and tailwind.

Challenges we ran into

Producing consistent results from the agent collaboration was difficult. The agents could not consistently decide when to read or write information to the database.

Accomplishments that we're proud of

We were proud of developing the main structure of the app and began utilizing a basic agentic workflow to guide a user's prompt through the lesson generation and knowledge graph generation. Orchestrating different agents to collaborate on designing a lesson plan, using the information in the plan to generate a knowledge graph personalized to a users learning journey.

What we learned

We learned a lot about how different AI Agents work together to improve reasoning capabilities and enhance results. Prompting and providing instructions for each agent to produce consistent, desirable results is a uniquely difficult task.

What's next for Knowde

During this hackathon, we were able to build a strong starting point for modular improvements in the future. We would love to enhance the personalization in updating a user's knowledge graph, utilize multimodal inputs to minimize interruptions in a study block, and implement relevant search history to adapt a user's learning plan.