Mirror Mind

login page
landing page
main ui
knowledge graph
synced notion data
generated POC PR
notification email

Inspiration

People often get ideas for products or solutions to problems they or someone they know experienced. But the timing is rarely ideal to pull out a phone and start typing. Many ideas appear while walking, commuting, or doing something else. By the time someone sits down to write them, parts of the idea are already lost.

We wanted to solve this problem by creating a system that can capture ideas the moment they appear and help shape them while they are still forming.

What it does

Mirror Mind is an AI thinking companion that listens to spoken ideas and turns them into structured thoughts in real time.

Users can simply talk naturally, like speaking to a friend. The system captures fragmented thoughts and organizes them into a live concept map and structured thought tree.

As the idea develops, the agent asks clarifying questions to refine the concept, evaluate feasibility, and guide the thinking process. A short burst of messy thoughts can quickly evolve into a clear idea, structured plan, or next set of actions.

How we built it

We built the prototype using Gemini Live for real time audio understanding and streaming interaction.

A backend service runs on Cloud Run to manage agent orchestration and system logic. Redis is used to maintain conversational state and store the evolving idea structure.

For deeper reasoning and expansion, requests are sent to Vertex AI models to evaluate feasibility, generate structured outputs, and refine the idea tree.

The resulting knowledge structure can also be exported to tools like Notion so ideas can immediately become notes, tasks, or early product documentation.

Challenges we ran into

Audio cutoff during streaming was one of the first challenges. Handling real time speech while maintaining context required careful buffering and segmentation.

We also faced resource exhaustion during heavy model usage when several inference steps occurred quickly.

Another challenge was integration and maintaining the evolving idea graph in an abstract structure while still allowing users to pivot ideas and restructure them naturally.

Accomplishments that we're proud of

We built a system that can transform a short, unstructured spoken idea into a structured concept map and thought tree within seconds.

The ability to interrupt the agent, pivot the idea, and see the structure update in real time creates a natural interaction that feels closer to thinking out loud than prompting an AI.

We are proud that the system helps turn messy thinking into something that can evolve into a product idea or actionable plan.

What we learned

We learned that human thinking is rarely linear. People explore ideas through fragments, tangents, and partial concepts.

Designing AI systems to support this process requires flexible structures that can evolve and reorganize dynamically.

We also learned that real time interaction fundamentally changes how people collaborate with AI compared to traditional prompt based systems.

What's next for Mirror Mind

Next we plan to improve the concept mapping interface and make it fully interactive through voice and gesture controls.

We also want to deepen integrations with productivity tools so ideas can automatically transform into tasks, documentation, or early prototypes.

Our long term goal is for Mirror Mind to become a thinking companion that helps people move from raw thoughts to real products faster.

it will also research other existing applications that are or might be related to your idea and provide you with pros and cons of it and how you could improve your idea.