Remi

We remember, so you don’t have to. Stop managing your work. Start doing it

Comment

Inspiration

Modern knowledge workers are drowning in digital chaos. The average professional switches contexts 300+ times daily, loses 2.5 hours to inbox management, and spends 31 hours monthly in unproductive meetings.

  • Email Overload : 147+ messages per day demanding immediate attention

  • Meeting Fatigue: Back-to-back calls with no time to process or act

  • Context Switching: Constant interruptions fragmenting deep work

What it does

  • Smart Inbox: Automatically triages, summarizes, and drafts responses to emails based on context and priority.

  • Live Meeting Agent: Joins calls, captures decisions, assigns action items, and updates your systems in real-time

  • Daily Briefing: Say, "Remi, what does my day look like?" Wake up to a personalized overview: what matters today, why it matters, and what success looks like

How we built it

We built Remi as a full-stack intelligent orchestration pipeline:

  • Frontend: A desktop application built in Electron-JS (TS)
  • Backend: Python (Flask) orchestrator coordinating multiple “agents” (email_agent, meeting_agent, daily_agent)
  • Integrations:
    • Gmail API → fetch + summarize unread emails, as well as automate sending emails using a Web Service deployed on Render
  • Google Calendar API → sync daily meetings
  • Supabase → unified database for storing parsed events, summaries, and AI-generated replies Photon iMessage Kit → deliver the daily briefing directly to the user’s Messages app
  • AI Core: Gemini 2.5 and custom llm_client for contextual summarization, sentiment analysis, and natural reply drafting
  • Database schema: unified emails and events tables with fields for summaries, action items, sentiment, and suggested responses
  • Workflow orchestration: orchestrator.py triggers each agent sequentially (emails → meetings → daily digest → iMessage delivery)
  • Data safety: Each item is stored once per day with message deduplication (message_id / timestamps)

Challenges we ran into

  • Real-time audio sync: Getting live meeting transcription to stay synchronized with AI processing and frontend updates without lag or dropped context
  • Knowledge graph architecture: Building a context graph that dynamically connects emails, meetings, and people while maintaining query performance as relationships scaled
  • Multimodal integration: Coordinating Gmail API, Calendar sync, live audio streams, and AI inference in a single coherent pipeline without bottlenecks

Accomplishments that we're proud of

  • Real-time meeting intelligence - Live transcription feeding AI analysis that surfaces action items and context during the call, not after
  • Zero-friction delivery - Users wake up to an iMessage briefing that saves 2+ hours before they even open their laptop - no app install, no learning curve, just immediate value
  • Remi isn’t just another productivity app - it’s the start of a new workflow where your AI agent works as your chief of staff, running your day, filtering your noise, and freeing your time.

What we learned

  • We learned how powerful clean data flow and smart orchestration are when building autonomous AI agents.
  • Integrating Gmail, Calendar, and Supabase taught us to design scalable pipelines with proper deduplication and schema design.
  • Working with Photon’s iMessage Kit showed that great AI is about timing and delivery — not just intelligence.
  • Most importantly, we realized true productivity comes from AI that quietly works for you, not more apps or dashboards.

What's next for Remi

Next, we plan to integrate Slack and Google Tasks, add voice-based briefings, and expand Remi into a fully autonomous workplace agent that plans, responds, and executes - all before you open your laptop

Built With

Share this project:

Updates