SignalGarden

Main Screen
Selection Screen
What if Screen - after scenario is ran

🌱 Inspiration

Modern problems are no longer simple. Founders, researchers, and social-good teams often work inside messy systems made of people, goals, risks, ethics, and feedback loops. Most AI tools answer questions, but they don’t help us understand systems.

SignalGarden was inspired by the idea that insight doesn’t come from a single answer, but from seeing how forces interact over time. We wanted to build something that feels less like a chatbot and more like a living map of meaning.

🧠 What It Does

SignalGarden transforms unstructured inputs like notes, conversations, and whiteboard images into a living system graph.

Using Gemini 3’s multimodal reasoning, it:

Extracts entities, goals, risks, and feedback loops
Builds an evolving system graph
Simulates “what if” changes
Explains outcomes in clear, human language

Instead of static answers, users explore how systems behave, change, and break.

🛠 How We Built It

SignalGarden is built around Gemini 3 as a stateful reasoning engine, not just a response generator.

Core flow:

User submits text or images
Gemini 3 performs multimodal extraction
Semantic elements are structured into a graph
The current system state is preserved and reused
New inputs or simulations mutate the system
Gemini explains why changes occur

Gemini’s long-context reasoning allows the system to evolve instead of resetting on every prompt.

🤖 Gemini 3 Integration

Gemini 3 is central to SignalGarden’s functionality:

Multimodal reasoning combines text and images into one reasoning space
Long-context understanding maintains system history
Explainable outputs narrate structural changes and impacts
Low-latency inference enables interactive exploration

Gemini is not an assistant here. It is the gardener of the system.

🚧 Challenges

The biggest challenge was designing prompts and schemas that allow Gemini to reason structurally rather than conversationally. Another challenge was balancing abstraction with clarity so insights feel powerful but understandable.

📚 What We Learned

We learned that when AI is used to model systems instead of conversations, it unlocks a new class of understanding. Multimodal reasoning becomes far more powerful when paired with persistent state and visual representation.