PathWise

Inspiration

We were inspired by the gamified revision loop and clear roadmap that Duolingo use to make learning languages addictive. Agentic AI is now capable of building personalised courses for anything you can think of, and we have provided the agents with a platform to make these visually engaging and interactive.

What it does

Upload revision documents and textbooks, or briefly describe what you want to learn. You can highlight key areas you don't want us to miss, or take a back seat and focus on the recall. We generate a knowledge bank from your materials and provide a suite of AI agents and tooling to maximise your efficiency.

Our personalised AI tutor gathers insights from your quiz performance and previous sessions to identify your weak points, explain anything you don't understand and subtly push you in the right direction.

Our gamified quiz roadmap is a low-pressure, fun way to study your topic that you can pick up anywhere. The quiz uses proven active recall and spaced repetition principles to ensure learning is fun and productive.

How we built it

Our Next.js frontend is connected to a minimal Firestore for persistent data storage. When a user submits a prompt to generate a roadmap, we use Claude-Haiku with LangChain to perform deep research on the web, break down the key topics sorted by complexity, generate questions for each section and organise this data for the frontend. The LangChain agent also gathers insights based on quiz performance and relays this to the personalised AI tutor.

Challenges we ran into

One of our main challenges was transforming a technically complex multi-agent system into an intuitive and engaging user experience.

Accomplishments that we're proud of

The utility of the personalised AI tutor, which moves beyond simple question answering to actively guide learning

What we learned

This was our first exposure to orchestrating AI agents. We learned how to design agent workflows, manage context sharing between models, and build feedback loops driven by user interaction data.