Inspiration

I noticed a fundamental flaw in how current wellness technology interacts with us. Wearables today are essentially "anxiety trackers." They monitor your stress, and when you hit a breaking point, they buzz your wrist with a notification asking, "Are you okay?" But by the time you get that notification, you're already spiraling. Worse, they lack context. If your heart rate spikes because you're watching a thriller on TV or going for a run, the app still panics and sends you a mindfulness prompt.

I wanted to build something different. I asked myself: What if technology didn't just track our stress, but actively absorbed it? What if our homes could act as an external nervous system, protecting us before we even consciously realize we're overwhelmed? That question became Project Resonance.

What it does

Project Resonance is a context-aware, ambient cognitive infrastructure. It turns a smart home into a closed-loop wellness sanctuary that operates entirely in the background.

When a user's biometric data indicates rising physiological arousal (the precursor to anxiety or sensory overload), Resonance doesn't just send an alert. Instead, it pauses and reasons. It checks the environment: Is the user watching an action movie? Are they exercising? Or are they sitting quietly at their desk at 11 PM?

Only when the system confirms the stress is negative cognitive overload does it gently intervene. It autonomously dims the smart lighting, mutes external distractions like the doorbell, and transitions the living room display into a bio-reactive, ambient visual that subtly guides the user's breathing back to baseline. It requires zero taps, zero apps, and zero friction.

How I built it

At its core, Resonance is a self-hosted Alexa+ MCP (Model Context Protocol) Server built using the Streamable HTTP specification. I chose Alexa+ as our primary track because the MCP architecture is the perfect "brain" for orchestrating a multi-device ecosystem.

I built the system using a strict event-driven architecture:

• The Senses: I created a simulated biometric webhook (representing the Bee wearable) that sends anonymized "Arousal State Tokens" to our MCP server.
• The Brain: I used Amazon Bedrock to power our agentic loop. I utilized Claude 3.5 Sonnet for the core reasoning and tool-chaining logic, and Amazon Nova for multimodal vision to analyze environmental context from a simulated Ring camera snapshot.
• The Actuation: Once the agent reasoned that intervention was necessary, it triggered a custom tool to push ambient protocols to a simulated Fire TV environment.
• The Glass Box: To make the invisible AI visible, I built a real-time Streamlit dashboard that visualizes the agent's "Chain of Thought," showing judges exactly how the AI fuses biometric and visual data to make a decision.

Challenges we ran into

The biggest challenge wasn't technical; it was philosophical. How do I prevent false positives without invading the user's privacy? If the AI needs to "see" the room to know if the user is exercising, does that mean we are constantly streaming video to the cloud?

I solved this by architecting a Zero-Knowledge Context Fusion model. Raw biometric data and raw video never leave the local edge. The Ring camera extracts a simple, anonymized metadata tag (e.g., {"scene": "sedentary"}), and the wearable sends a simple state token (e.g., {"state": "elevated"}). The Alexa+ MCP server only ever sees these anonymous tokens. This ensured the system is inherently privacy-preserving and compliant with strict data regulations, while still being highly intelligent.

Accomplishments that we're proud of

I am incredibly proud of the "restraint" I programmed into the AI. During testing, I fed the system a scenario where the biometric data showed extreme stress, but the visual context showed the user running on a treadmill.

Also proud of our "Glass Box" Streamlit UI. Making agentic reasoning transparent and beautiful is incredibly difficult, but I created a dashboard that makes the AI's thought process feel intuitive and trustworthy.

What I learned

I learned that true ambient intelligence requires profound restraint. The temptation when building AI agents is to make them do more. But in wellness and cognitive care, the most powerful thing an AI can do is know when to act, and when to stay silent.

I also gained deep, hands-on experience with the new Alexa+ MCP Streamable HTTP spec and Amazon Nova's multimodal capabilities. I learned how to structure tool schemas so that an LLM can reliably chain complex, multi-step environmental commands without hallucinating.

What's next for Resonance

While I built this as a consumer wellness tool, we see massive potential in B2B2C applications. I plan to pilot Resonance as a corporate benefit for remote workers in high-stress industries (like healthcare and finance) to prevent executive burnout. I also see a profound use case in "aging-in-place" care, helping early-stage dementia patients feel grounded and safe in their homes.

Finally, I am open-sourcing our resonance-mcp-toolkit. I want to give the broader developer community a privacy-first, pre-built template so they can easily add environmental context to their own Alexa+ wellness agents. I want to set a new standard for empathetic, ambient computing.

Built With

• accessibility
• artificial
• fastapi
• health
• home
• intelligence
• pydantic
• python
• smart
• streamlit
• wellness