I built an Agentic Calendar & Meeting Assistant that uses AI to understand user requests and interact with calendar workflows.
The assistant can list and manage calendar events; create new meetings through natural language; reschedule existing events; cancel meetings.
The assistant maintains conversational context using memory and streams AI responses in real time.
My frontend stack is Next.js, React, TypeScript, Tailwind CSS; my backend is Node.js, Express.
My agent and data stack uses Mastra, Google Gemini, PostgreSQL, LibSQL memory, and Google Calendar API.
I use Server-Sent Events (SSE), Zod, and Descope for streaming, validation, and authentication.
The engineering challenges include converting natural-language requests into structured actions; managing conversation context and memory; handling API failures and retries; validating inputs before calendar operations; maintaining reliable communication across the frontend, backend, and agent.
My biggest takeaway is that an effective AI agent needs reliable tools, controlled execution, memory, validation, and proper error handling—not just a good prompt.
This project helped me explore how LLMs can move beyond answering questions and start performing useful real-world tasks.
I shared a live demo of the calendar assistant at https://lnkd.in/gUSUwV4w.