A full-stack AI agent that classifies incoming Gmail messages with GPT-4, drafts smart replies, and auto-schedules meetings via Google Calendar — with an approval-gated Next.js dashboard for human-in-the-loop review

Manually triaging email, drafting replies, and coordinating meeting times is repetitive and time-consuming, and naive polling-based automation approaches waste LLM tokens by reprocessing old or already-seen messages.
Built a multi-service architecture — a Next.js dashboard plus a separate Node worker — that uses Gmail's historyId-based incremental sync to process only new emails, GPT-4 structured outputs for classification and meeting-intent extraction, and the Google Calendar API for free/busy checks and event creation, all gated behind an approve/reject action feed.
Classifies incoming emails by category (meeting request, urgent, newsletter, spam) using GPT-4 structured outputs.
Extracts proposed meeting times, checks Google Calendar availability, and prepares event creation.
Generates contextual reply drafts for review before sending, never auto-sends.
Reduced redundant LLM calls by switching to Gmail historyId-based incremental sync instead of full inbox polling.
Automated meeting scheduling by connecting GPT-4 meeting-intent extraction directly to Google Calendar free/busy checks and event creation.
Built an approval-gated dashboard so every AI-drafted reply and calendar event is reviewed before it goes live.