Lumi: AI Life & Work Companion

Lumi is a multi-persona AI companion for modern knowledge workers and founders. From a single interface, Lumi can switch between three "personas": Lumi Lead, Care, and Shop.

Comment

⭐ Inspiration

Modern work and modern life are both overwhelming — yet the tools meant to help us are fragmented. Engineering leaders jump between PRs, incidents, Slack messages, and Jira updates. People struggle silently with stress, burnout, and decision fatigue. Shopping for the right product still takes hours of browsing.

We asked a simple question:

What if a single AI companion could support you across your work, your wellbeing, and your daily decisions — without switching apps?

This led to Lumi: an AI Life & Work Companion that adapts to your needs through three personas: Lumi Lead, Lumi Care, and Lumi Shop.

⭐ What it does

Lumi provides three tightly integrated AI experiences:

1. Lumi Lead — Engineering & Startup Assistant

  • Summarizes PRs and highlights risk
  • Generates daily engineering briefings
  • Prepares 1:1 talking points
  • Predicts sprint velocity and workload health

2. Lumi Care — Voice-Based Emotional Support

  • Users talk freely about their stress, goals, or dilemmas
  • Lumi responds in a warm “Future You” voice via ElevenLabs
  • Tracks emotional patterns over time
  • Provides reflection cards and wellbeing guidance

3. Lumi Shop — Agentic Shopping Assistant

  • Understands shopping intent (gift, outfit, budget)
  • Searches catalogs and produces curated recommendations
  • Performs price comparisons using Vultr compute
  • Builds personalized profiles through SmartMemory

Across all personas, Lumi maintains a unified identity, enabling deep personalization with zero friction.

⭐ How we built it

Backend on Raindrop MCP Server

  • Multi-agent architecture built with SmartInference workflows
  • Persona-specific memory domains using SmartMemory
  • Structured analytics using SmartSQL
  • File ingestion (PR diffs, logs, catalogs) through SmartBuckets

Ultra-Low Latency Processing

  • Cerebras Inference for emotional tone detection, PR triage, and rapid shopping scoring

Voice Experience

  • ElevenLabs STT + TTS for Lumi Care’s “future you” conversational style

Frontend

  • Next.js + React with a clean persona switcher
  • Real-time voice recording and playback
  • Dynamic briefings, mood charts, and shopping results

Infrastructure

  • Vultr Compute for scraping, price analysis, and catalog processing
  • Vultr Object Storage for datasets and logs

The result: three AI experiences, one unified system, one seamless UI.

⭐ Challenges we ran into

  • Building three distinct agents with one shared identity Designing SmartMemory boundaries and cross-persona context without leakage required thoughtful architecture.

  • Achieving the “instant response” feeling Integrating Cerebras for low-latency emotional and PR analysis required tuning and careful workflow design.

  • Voice UX matters more than expected Getting Lumi Care to feel warm, supportive, and natural involved prompt tuning, emotional context retention, and smoothing ElevenLabs output.

  • Balancing scope vs. simplicity Lumi touches leadership, wellness, and shopping — deciding what belonged in v1 vs. future versions took discipline.

⭐ Accomplishments that we’re proud of

  • Built a unified multi-persona AI companion that feels cohesive, not stitched together
  • Achieved ultra-fast emotional inference with Cerebras
  • Designed a voice-first mental wellbeing flow that feels comforting and human
  • Integrated Raindrop’s SmartComponents cleanly to power personalization, analytics, and memories
  • Delivered three high-value use cases within one interface — without overwhelming the user

⭐ What we learned

  • Multi-agent apps shine when they share memory, not prompts
  • Voice UX requires more than STT/TTS — it requires emotional alignment
  • SmartInference workflows make it possible to orchestrate domain-specific reasoning at scale
  • Low-latency inference fundamentally changes user perception of “AI helpfulness”
  • Raindrop + Vultr + Cerebras + ElevenLabs is a powerful stack for agentic apps

⭐ What’s next for Lumi: AI Life & Work Companion

  • Deeper integrations

    • GitHub
    • Jira
    • Slack
    • Amazon/Shopify shopping APIs

  • Mobile app with always-on voice

  • Personal OS features: calendar, email triage, meeting prep

  • Multi-modal PR analysis with UML diagrams and architecture suggestions

  • Emotion trend prediction and personalized wellbeing plans

  • Shopping negotiation agent that automatically contacts sellers for offers

  • Team dashboard for Lumi Lead (velocity, risks, contribution heatmaps)

Our vision: Lumi becomes a personal AI operating system that supports your entire life.

⭐ Built with

Languages & Frameworks

  • TypeScript
  • Node.js
  • Next.js
  • React
  • TailwindCSS

AI & Cloud Platforms

  • Raindrop Platform (MCP Server, SmartInference, SmartMemory, SmartSQL, SmartBuckets)
  • Cerebras Inference (low-latency emotional + PR + shopping scoring)

  • Vultr Cloud

    • Compute Instance (scraping, heavy tasks)
    • Object Storage (datasets, logs)
    • Optional GPU workloads

Voice Technologies

  • ElevenLabs (voice generation + optional STT)

Databases & Storage

  • SmartSQL (structured data)
  • SmartBuckets (files, catalogs, diffs)
  • Vultr Object Storage

Other Tools

  • WorkOS (optional authentication)
  • Vercel / Netlify (frontend deployment)

Built With

Share this project:

Updates