AI / ML · Software Developer
Canvas MCP
AI-Powered Academic Productivity Assistant
An AI-powered assistant that unifies Canvas LMS, Google Calendar, and Gmail into a single natural-language chat interface, built with FastAPI, Streamlit, and Claude (via OpenRouter) using the Model Context Protocol (MCP).
Highlights
- 01
Unified 3 live data sources (Canvas LMS, Google Calendar, Gmail) behind a single chat interface by architecting a FastAPI service layer with 87 MCP-registered tools, enabling users to query assignments, schedule events, and manage email through plain English with zero context switching.
- 02
Eliminated temporal hallucinations on all relative-time queries by injecting the live server-side date, time, and IANA timezone into every system prompt at request time, reducing incorrect date inferences to zero and enabling reliable scheduling commands like "create a study session for tomorrow at 3 PM."
- 03
Built a fully agentic multi-step reasoning loop — supporting up to 10 sequential tool-call iterations per query — allowing the AI to automatically chain tools (e.g., fetch courses → fetch assignments → create calendar block) without additional user input, surfacing each invocation in a real-time transparency panel.
- 04
Deployed a full-stack AI application (FastAPI backend on Render, Streamlit frontend on Streamlit Cloud) featuring OAuth 2.0 Google integration, environment-configurable model selection across OpenRouter providers, and a structured system prompt with explicit behavioral rules, serving as a production-ready productivity platform for students.