Lumen

Home page
End-of-Day Summary
Chatbot
Chatbot query
Chatbot query

Inspiration

Students and professionals today juggle multiple productivity platforms like Gmail, Google Classroom, and Google Calendar. Important deadlines, meetings, and emails often get buried under notifications, causing missed assignments and overlooked opportunities.

We wanted to build something that doesn’t just show notifications, but actually understands a user’s workspace and helps them prioritize tasks intelligently. That idea led to Lumen — a smart workspace assistant that observes, summarizes, and communicates what truly matters.

What it does

Lumen is an AI-powered workspace agent that connects with Gmail, Google Classroom, and Google Calendar to provide intelligent insights and assistance.

It can:

📧 Analyze emails and highlight important or urgent messages
📚 Track assignments and deadlines from Google Classroom
📅 Monitor meetings and schedules from Google Calendar
🧠 Generate automated End-of-Day (EOD) productivity summaries
💬 Provide a chat interface where users can ask questions like:
- “What do I have due this week?”
- “Show my unread emails”
- “Do I have any meetings today?”

Lumen uses AI reasoning to convert raw workspace data into natural language insights, helping users stay organized without manually checking multiple platforms.

How we built it

We built Lumen using a modular architecture:

Backend: FastAPI (Python) for API endpoints and agent orchestration
AI Reasoning: Google Gemini API for summarization, chat responses, and workspace analysis
Data Sources:
- Gmail API for email observations
- Google Classroom API for assignments and courses
- Google Calendar API for meetings and scheduling
Database: SQLAlchemy with SQLite to store:
- Daily workspace snapshots
- Chat history
- Generated productivity reports
Autonomous Agent System:
- Periodically collects workspace data
- Generates insights and EOD reports automatically
- Uses structured JSON outputs for reliable reasoning

The system uses intent routing to determine whether a user query relates to emails, meetings, assignments, or general workspace assistance.

Challenges we ran into

One major challenge was managing AI reliability and structured output parsing. Language models sometimes return unstructured text, which caused errors when converting responses into structured insights.

Another challenge was handling API failures and quota limits. We implemented fallback logic so Lumen can still respond using stored workspace data even if the AI model becomes unavailable.

We also faced challenges in intent detection and context memory, ensuring the assistant correctly interprets user questions and remembers previously referenced tasks or meetings.

Finally, integrating multiple Google APIs while maintaining synchronization and avoiding redundant data storage required careful system design.

Accomplishments that we're proud of

We successfully built a fully working autonomous workspace agent that:

Integrates multiple productivity platforms into one intelligent system
Generates automated productivity summaries without user input
Supports natural language queries across workspace data
Implements fallback and error-resistant AI reasoning
Maintains structured daily snapshots for consistent analysis

We are especially proud of designing Lumen to function as a proactive assistant, not just a reactive chatbot.

What we learned

Through this project, we learned:

How to design multi-source AI agents using real-world APIs
Techniques for enforcing structured JSON outputs from language models
How to implement fault-tolerant AI systems with fallback logic
The importance of intent routing and contextual memory in conversational agents
Best practices for scalable backend architecture using FastAPI and asynchronous workflows

We also gained deeper insight into building AI systems that are reliable, user-centric, and production-ready.