HomeSight

Inspiration

Caring for aging family members from a distance creates constant background anxiety. Standard smart home systems are reactive—they chime when a motion sensor triggers or when a doorbell rings—but they lack context. They can't answer nuanced questions like, "Has Dad been unusually quiet this morning?" or "Did his afternoon medicine reminder get acknowledged?"

With the rollout of Alexa+ and its open Model Context Protocol (MCP) framework, we realized we could bridge the gap between static smart home devices and intelligent, proactive caretaking. We were inspired to build HomeSight: a contextual eldercare agent that turns scattered home events into meaningful, natural-language safety insights for families.

What it does

HomeSight is an intelligent Alexa+ care companion that connects family members to an aging loved sister or parent's daily routine using natural conversation.

Contextual Safety Check-Ins: Users can ask Alexa+, "Check on Mom's daily routine," and HomeSight synthesizes multi-sensor inputs to summarize her morning activity, mobility patterns, and environmental safety.
Proactive Routine Monitoring: If routine anomalies occur—such as no activity in the kitchen during breakfast hours or an unacknowledged medication window—HomeSight generates structured alerts and suggests appropriate follow-ups.
Agentic Escalation: In critical scenarios, the agent can coordinate sub-tasks, such as sending an urgent summary to a primary caregiver's phone or preparing a timeline for emergency responders.

How we built it

HomeSight is architected around the Alexa+ MCP ecosystem and powered by AWS cloud infrastructure:

Alexa+ Integration: Built a self-hosted MCP Server adhering to the 2025-11-25 specification over Streamable HTTP. This exposes structured tools and contextual resources directly to Alexa+.
AWS & Agent Core: Deployed on AWS using Amazon Bedrock for high-reasoning tasks and Strands SDK to coordinate specialized sub-agents (e.g., activity pattern analysis vs. alert escalation).
Developer Tools: Developed using Kiro Crew for rapid multi-agent orchestrations, allowing clean state management and event processing.
Open Source Framework: Released the core Alexa+ MCP integration layer as an open-source template to help other developers build agentic tools for Alexa+.

Challenges we ran into

MCP Streamable HTTP Setup: Implementing the late-2025 MCP spec for Streamable HTTP required strict adherence to session handling and chunked responses, demanding precise lower-level network debugging.
Context vs. Noise: Smart homes generate high volumes of low-level telemetry. Filtering raw sensor logs into high-level, human-friendly summaries required rigorous prompt-engineering and agentic routing via Amazon Bedrock.
Low Latency Voice Responses: Ensuring the MCP tool-calling pipeline executed within conversational response limits required optimizing our AWS Lambda and Bedrock orchestration steps.

Accomplishments that we're proud of

Seamless MCP Integration: Successfully launched a fully spec-compliant, self-hosted MCP server that allows Alexa+ to execute multi-step reasoning workflows cleanly.
Privacy-First Design: Engineered the system to process ambient routine data without recording unnecessary visual or audio telemetry, focusing strictly on structured event signals.
Reusable Open-Source Contribution: Created a clean, well-documented boiler-plate repository for developers looking to connect AWS Bedrock workflows to Alexa+ via MCP.

What we learned

The Power of Open MCP Standards: Leveraging open standards like MCP dramatically simplifies connecting complex cloud AI models to consumer interfaces like Alexa+.
Contextual AI in Eldercare: We learned that caregivers don't want raw data logs; they want synthesis, reassurance, and actionable answers to plain-language questions.

What's next for HomeSight

Multi-Device Sensor Fusion: Expanding the MCP resource endpoints to ingest real-time streams from Ring doorbells and Bee wearable devices for richer movement tracking.
Custom Care Plan Setup: Allowing caregivers to configure custom routine goals and threshold parameters directly through a companion web dashboard.
Offline Fallback Routing: Implementing edge-cached agent behaviors to ensure local alerts remain operational even during internet outages.