MeetWise

An autonomous AI meeting assistant built with the AWS Strands Agents SDK that schedules meetings, prepares agendas, and sends summaries — with a human approval gate for safety.

Comment

Inspiration

Every week, professionals lose hours to the same repetitive post-meeting ritual: writing notes, extracting action items, and emailing summaries. For a manager in five meetings a week, that's four or more hours of lost productivity — time that should go to real work. I built MeetWise to change that.

What It Does

MeetWise is an autonomous AI agent built with the AWS Strands Agents SDK. It handles the entire post-meeting workflow:

  1. Schedules meetings based on topic, attendees, and preferences
  2. Prepares structured agendas automatically
  3. Drafts and sends summaries — but only after explicit user approval

The agent runs quietly and only surfaces when there's a real decision to make.

How I Built It

MeetWise uses three @tool decorated functions:

  • schedule_meeting — suggests a meeting slot based on topic and preferred time
  • prepare_agenda — generates a structured agenda with all attendees
  • send_summary — sends the summary, with a built-in Approval Gate

The agent runs on Amazon Bedrock using the Amazon Nova Pro model, keeping it fast and cost-effective.

The Approval Gate — Why It Matters

The most important design choice: every irreversible action requires explicit human approval. Before sending anything, the agent shows the exact content and asks for a yes/no confirmation. This keeps the human in control and makes the agent safe for real professional use.

This pattern is critical for any agent that touches real-world systems — email, calendars, payments. Users need to trust that the agent won't act without them.

Challenges I Faced

  • Retired model IDs on Bedrock — the original Claude 3.5 Sonnet model I started with had reached end-of-life. I migrated to Amazon Nova Pro to keep the agent running without payment-instrument issues.
  • Building a conversational loop on top of the Strands SDK so the user could have a multi-turn dialogue with the agent.
  • Implementing a tool-level Approval Gate that pauses execution cleanly within the tool call.

What I Learned

The Strands Agents SDK makes tool-driven agent design remarkably clean. The @tool decorator and the automatic LLM reasoning loop let me go from idea to a working agent in a single day. The Approval Gate pattern is essential for professional-grade agents.

What's Next

  • Google Calendar and Gmail integration
  • Meeting transcript ingestion to auto-generate summaries
  • Multi-user team workspaces
  • Deploy on AWS AgentCore for production scale

Built With

  • Strands Agents SDK (Python)
  • Amazon Bedrock (Amazon Nova Pro)
  • AWS (us-east-1)

Built With

Share this project:

Updates

Submission history