Cornell Personal Assistant

Overview

I’m building a personal assistant system inspired by multi-agent orchestration patterns. The Core Orchestrator interprets natural language, routes tasks to specialized agents, and stores long-term memory in PostgreSQL.

Problem

Managing classes, deadlines, reminders, job search tasks, and workflows across tools (Calendar, email, notes, trackers) creates constant context switching and missed tasks.

Solution

A Core Orchestrator Agent that:

• Extracts intent and entities from user requests
• Routes to specialized agents (Calendar, Email, Finance, Weather & News, Research, Content Creation)
• Stores and retrieves memory from PostgreSQL
• Generates structured outputs that sync into tools like Notion

Architecture

Flow

1. User request → Core Orchestrator (GPT)
2. Orchestrator outputs { intent, entities, next_agent }
3. Specialized agent runs actions (API calls, summaries, plans)
4. Results are saved to memory (PostgreSQL) and pushed to downstream tools (Notion)

Key design choices

• Modular agents so each workflow remains maintainable
• Strong prompt contracts for predictable JSON outputs
• Memory layer for continuity across sessions

Tech Stack

• OpenAI API (routing + reasoning)
• n8n (workflow automation)
• PostgreSQL (memory)
• Google Calendar
• Notion

Results

• Reduced manual weekly planning overhead
• More consistent follow-through via structured action plans

What’s next

• Email triage agent
• Weather & News daily digest
• Resume and job-search automations