SlugOps

Inspiration

I built SlugOps because campus life is fragmented across calendars, dining menus, transit, reminders, weather, and news. I wanted a grounded campus agent that could help plan a student’s day using real data instead of guessing.

## What it does

SlugOps creates daily campus plans using Google Calendar, UCSC dining hall data, transit estimates, weather, reminders, preferences, and news. It supports morning insights, dining retrieval, calendar event creation, recurring events, duplicate prevention, and source-linked news summaries.

## How we built it

I built SlugOps as a local MCP-powered agent tool system using Python, SQLite, Google Calendar, UCSC dining data, RAG-style retrieval, RSS news search, transit data, and structured tool grounding. The agent reads from real tools and returns source-aware outputs for safer planning.

## Challenges we ran into

The biggest challenge was preventing hallucinations. I had to make calendar data real-only, surface tool failures clearly, avoid fake fallback events, protect preferences, require confirmation for writes, and stop duplicate calendar events before they happen.

## Accomplishments that we're proud of

I built a working campus operations agent with grounded daily planning, real Google Calendar reads/writes, recurring calendar events, duplicate protection, SQLite/RAG dining search, safe preference updates, news summaries with links, and 150+ passing tests.

## What we learned

I learned that useful agents need more than strong prompts. They need reliable tools, explicit source grounding, safe write boundaries, failure handling, and tests. The hard part is not making the agent sound smart. It is making sure every claim traces back to real data.

## What's next for SlugOps

Next, I want to improve the chat experience, expand dining nutrition intelligence, add richer calendar workflows, improve transit reliability, support more personalized morning insights, and reorganize the architecture into cleaner domains for long-term development.