-
-
Meeting Prep Assistant — simulated Alexa+ experience
Inspiration
Before any meeting, I always found myself digging through old notes, emails, and calendar invites just to remember who I was talking to and what was still unresolved. I wanted something that could do that digging for me out loud, like a real assistant.
What it does
Meeting Prep Assistant is a voice-first assistant that briefs you before a call. Ask it to prep you for a meeting, and it pulls together who's attending, their role, the last thing that happened with them, and any open issues tied to the relationship then reads it back to you in plain, spoken language. You can ask natural follow-ups too, like "who else is joining?", and it remembers the context of the conversation instead of making you repeat yourself.
How I built it
This is built for the Alexa+ track using the hackathon's simulated-experience path, a web app with speech input and spoken output (via the Web Speech API) that behaves like a voice assistant, rather than a self-hosted MCP server. Under the hood:
- Node.js + Express backend
- Amazon Bedrock (Converse API, Claude Sonnet 4.6) generates the actual spoken responses, grounded only in the meeting data provided it's instructed not to invent facts it doesn't have
- A simple in-memory session store keeps track of conversation history per session, which is what lets follow-up questions work naturally
Challenges I ran into
Getting Bedrock model access configured correctly was the biggest challenge. Newer models like Claude Sonnet 4.6 require an inference profile ARN rather than a plain model ID, which isn't obvious from the model's own overview page. I also hit a PowerShell execution policy block on Windows during setup. Both are documented in my friction log.
What I learned
I went from having never used AWS or a terminal to standing up a full Bedrock-backed application including IAM users, access keys, inference profiles, and debugging real API errors.
What's next
Swapping the mock meeting data for a real calendar and CRM integration, and adding persistent (not just in-memory) session storage.
Log in or sign up for Devpost to join the conversation.