EchoPilot AI

Dashboard

💡 Inspiration

Modern professionals, operators, and developers are overwhelmed by fragmented workflows. Traditional AI chatbots only generate text in a void—they cannot remember personal routines across days, cannot autonomously schedule commitments, and have no connection to device ecosystems like Alexa+. Furthermore, when an AI tries to make consequential changes (such as rescheduling meetings or clearing tasks), users lack confidence because there is no human-in-the-loop verification.

We built EchoPilot AI to bridge the gap between conversational AI and autonomous operations: combining Amazon Bedrock, a self-hosted Alexa+ Model Context Protocol (MCP) server, and Amazon DynamoDB with strict human approval gates.

🚀 What It Does

EchoPilot AI features a Dual-Engine Architecture:

1. Ask AI (General Knowledge & Reasoning)

Powered by Amazon Bedrock Converse API (with Anthropic Claude 3.5 Sonnet).
Provides comprehensive, ChatGPT-quality answers across computer science algorithms, mathematical evaluation, science, history, and technical writing with syntax-highlighted code blocks.
Does not mutate databases prematurely.

2. Alexa+ Operations Agent (Autonomous Execution Loop)

Follows the autonomous operational lifecycle: $$\text{Remember} \longrightarrow \text{Understand} \longrightarrow \text{Plan} \longrightarrow \text{Request Approval} \longrightarrow \text{Act} \longrightarrow \text{Verify} \longrightarrow \text{Learn}$$

Self-Hosted Alexa+ MCP Server: Built to the official Spec 2025-11-25 with Streamable HTTP & SSE Transport (/mcp/stream, /mcp/tools/call, /mcp/tools/list).
Persistent Memory: Stores user routines and study preferences in Amazon DynamoDB (e.g., "Remember that I prefer morning focus from 9 to 11 AM").
Deterministic Daily Planner: Formulates conflict-free schedules protecting deep work blocks and fixed meetings: $$\text{Scheduled Time} = \sum (\text{Task Duration} + \text{Buffer}) \leq \text{Available Work Hours}$$
Human-in-the-Loop Safety: Consequential actions (e.g., rescheduling meetings, bulk task mutations) trigger interactive Approve / Deny cards.
Verified State Mutability: Every write operation is verified against DynamoDB before confirming completion.
Voice Integration: Integrated speech synthesis (Text-to-Speech) and microphone dictation for a simulated Alexa+ experience.

🛠️ How We Built It

Strictly 100% Java 25 on Backend (Zero Python): Spring Boot 3.4.3, Jakarta EE, and AWS SDK v2 for Java (software.amazon.awssdk).
Amazon Bedrock Converse API: BedrockRuntimeClient using Claude 3.5 Sonnet (anthropic.claude-3-5-sonnet-20241022-v2:0) with bounded tool-calling loops.
Self-Hosted MCP Server (Spec 2025-11-25): 7 verified tools (save_memory, search_memory, create_task, list_tasks, plan_day, update_task, verify_action) with Streamable HTTP.
Amazon DynamoDB: DynamoDbEnhancedClient utilizing partition keys (userId) for strict multi-tenant account isolation.
Frontend: React 19, TypeScript, Tailwind CSS, Vite with bidirectional proxying.

🧗 Challenges We Faced

MCP Spec 2025-11-25 Compliance: Upgrading from legacy MCP versions to the hackathon-mandated 2025-11-25 specification required implementing custom Streamable HTTP and SSE emitters in Spring Boot.
CORS & Preflight Handling: Web clients communicating with self-hosted MCP endpoints across ports required implementing wildcard origin patterns (addAllowedOriginPattern("*")) in Spring Security.
Deterministic Constraint Resolution: Blending soft routines (morning study preferences) with hard constraints (fixed meetings) required a specialized interval scheduling solver.

🎓 What We Learned

How the Model Context Protocol (MCP) standardizes agent tools across the Amazon device ecosystem.
How the Bedrock Converse API normalizes tool schemas into uniform message streams regardless of underlying foundation model.
How to enforce safety in autonomous agents through structured human approval gates.

🔮 What's Next for EchoPilot AI

Expanding MCP integrations into Fire TV viewing schedules and Ring smart home routines.
Wearable health sync with Bee AI to automatically insert rest breaks into daily schedules.