Know-Way

About the Project

Inspiration

We based know-way on the simple concept of "You can't know what you don't know."

When people are trying to learn something new on their own, they often face major roadblocks and get stuck completely because they don't know where to look next. Other times, they make great progress but accidentally skip crucial foundational topics along the way. They end up tripping themselves over down the line when they approach an advanced topic that required those skipped steps as a strict prerequisite. We wanted to build something to solve that loop.

How We Built It & Lessons Learned

Being a two-person team with very little prior agent AI experience, the learning curve was steep. We relied heavily on Gemini to help us understand the fundamentals of agent orchestration and to get our system up and running.

Through a lot of trial and error, we learned how to quickly prototype a functional AI pipeline from scratch and gained a lot of hands-on experience navigating the unpredictable nature of working with LLM agents.

Challenges Faced

By far, the most perplexing part of building this pipeline was getting the agents to actually behave the way we wanted them to. We quickly realized how fragile prompting can be; occasionally, even the prompts we regarded as the safest, most secure ways to guide the agents would fail miserably for no apparent reason.

We eventually managed to get around this issue by enforcing Pydantic schemas as much as possible to strictly shape the agent outputs. It gave us the structure we needed, though to be honest, it can still be a bit wonky at times.

Built With

• adk
• and?react-flow?to-visualize-the-interactive-learning-nodes.-data-persistence-and-cross-agent-context-alignment-are-handled-using?mongodb-atlas
• antigravity
• clerk
• fastapi
• gemini
• mongodb
• nextjs
• react
• reactflow
• tailwindcss
• utilizing?gemini-2.5-flash?and?pro?models-hosted-on?google-cloud-vertex-ai?for-natural-language-understanding-and-graph-generation.-the-web-application-is-built-on?fastapi?for-asynchronous-api-routing
• with-a-frontend-powered-by?react.js