MailMind - AI Email Assistant

Email actions
Productivity insights
Overview
Calendar integration
Email lists

About the project

MailMind is a Chrome extension that upgrades Gmail with AI-driven analysis, triage, composition, calendar conversion, and bulk organization—right inside the Gmail UI. The goal is simple: move from inbox to done faster without context switching.

Inspiration

Our team spends a big chunk of the day in email. Even with modern AI, we were constantly switching tabs, copying and pasting, and losing focus across tools. We wanted a Gmail-native assistant that analyzes threads, suggests next actions, writes high-quality replies, turns emails into calendar events, and cleans up in bulk—all without leaving the inbox.

What we built

AI Triage & Insights: response time stats, backlog overview, sender/topic clusters, priority suggestions.
AI Composition: context-aware drafting, rewriting, summarization, tone and length control.
One-click Calendar: extract time/date and create calendar events from emails.
Bulk Actions: star, archive, label at scale based on AI analysis.
Gmail-native UI: floating action button (📧) and compose-window integration.

How we built it

Frontend/Extension: React 18 + TypeScript + Tailwind + Vite; MV3 with content.js, background.js, manifest.json.
Local AI (when available): Chrome Prompt API & Writer API (Chrome 127+ with flags) for on-device inference to reduce latency and keep UX smooth.
Fallback AI: Gemini API as compatibility fallback when local APIs aren’t available.
Backend: Node.js + Express with routes for ai, auth, mail, calendar.
Google Integrations: Gmail API for thread access and actions; Calendar API for event creation; OAuth2 for authentication.
Data flow: incremental thread fetching → lightweight feature extraction → AI scoring → UI suggestions → user-confirmed actions with undo.

Challenges

Sensitive data in emails: Prefer on-device AI via Chrome APIs to limit exposure; when falling back to cloud models, limit scopes, minimize payloads, and require user-visible confirmations for higher-risk actions.
Gmail/Calendar quotas & performance: batched requests, cached metadata, debounced UI updates, and a queued executor for bulk operations.
Ambiguity in NER/time parsing: multi-pass parsing with heuristics and confirmation before calendar creation.
MV3 constraints: adapt to service-worker lifecycles, persist state carefully, and use message passing between content/background scripts.

What we learned

On-device AI UX > raw model power for productivity tools: low latency and inline placement beat marginal quality gains from larger remote models.
Design for reversibility: undo/preview flows dramatically increase confidence in bulk actions.
Explainability builds trust: show why an email is “high priority” (sender recency, thread length, deadline hints) instead of opaque scores.
Operational guardrails are essential: quotas, rate limiting, backoff, and error handling matter even in a weekend build.

Impact (thinking about it quantitatively)

Definitions: $$ N := \text{emails processed per day} $$ $$ t_b := \text{baseline time per email (min)} $$ $$ t_a := \text{assisted time per email (min)} $$ $$ p := \text{fraction handled via bulk actions} $$ $$ \Delta_{\mathrm{bulk}} := \text{extra time saved per bulk-handled email} $$

Estimated daily time saved:

$$ S \approx N\,(t_b - t_a) + p\,N\,\Delta_{\mathrm{bulk}} $$

What’s next

Smarter prioritization: combine temporal urgency, relationship strength, and intent signals; better multi-language support.
Task center: extract todos from emails and sync with popular task systems.
Rule builder: human-readable automation rules with explanations and a safe simulation mode.
Multi-platform support: Chrome/Edge/Firefox extensions with a shared core, desktop add-ins for popular clients, and mobile share-sheet actions; unified settings and account sync.