SIIHA-Gemini-Multiagent

• 

  SIIHA: A human-centered AI assistant that integrates Gemini API with Gmail, Calendar, and Notion—emotionally intelligent, not just smart

• 

  SIIHA(Gemini x ADK) Diagram

Inspiration

In the chaotic flow of remote work, emotional overload often goes unheard. I created SIIHA—System for Integrating Intelligent Human Assistants—as a bridge between empathy and execution. My goal was to build a personal AI assistant that not only understands emotional signals like burnout or resignation, but also helps take action—booking meetings, sending emails, and documenting tasks—just by talking to it.

What it does

SIIHA-Gemini-Multiagent is a lightweight, emotionally aware assistant built on Gemini-2.0-flash-001 on Google ADK. It interprets natural-language requests from users—such as scheduling meetings, sending emails, or logging tasks—and automatically dispatches the right agent to handle them.

It currently supports three integrated agents:

• Google Calendar – auto-schedules meetings with semantic time/date parsing
• Gmail – composes and sends emails based on intent
• Notion – logs each task and action into a structured database for traceability

SIIHA also includes a layer of emotional prompting that mirrors user tone and provides supportive feedback, bridging emotional intelligence with practical function calling.

How I built it

I built a local-first multi-agent system with:

• Google adk web as frontend UI
• Gemini-2.0-flash-001 as the reasoning core
• Google Calendar & Gmail & Notion APIs for action
• Intent classification logic based on few-shot prompting The system classifies natural language (e.g. “I’m burned out”) into high-level intents (e.g. leave / feedback / resignation) and routes the request to corresponding agents.

Challenges I ran into

• Integrating Gemini routing with multiple OAuth-secured APIs required careful token flow management and error handling
• Ensuring consistent emotional response generation within Gemini’s function-calling limitations was a unique challenge

Accomplishments that I'm proud of

• Created a functioning multi-agent routing architecture with Gemini as the central brain
• Successfully connected Google Calendar, Gmail, and Notion via OAuth2 workflows
• Designed a lightweight intent classifier to dispatch user queries to appropriate agents
• Embedded emotional few-shot prompting to respond with both practicality and empathy
• Completed this project within intense time pressure (solo-developed in <48 hours)

What I learned

This was my first attempt to combine Google's Gemini function calling, OAuth-based access to Gmail, Calendar, and Notion, and intent classification all in one personal system. I learned to design routing prompts that feel natural yet precise, balancing emotion detection with productivity.

What's next for SIIHA-Gemini-Multiagent

• Expand agent capabilities: integrate Google Drive, Docs, or HRM systems
• Transition from few-shot to contextual memory for sustained conversations
• Deploy a cloud-native version with user-level authentication and agent chaining
• Improve the intent routing logic with dynamic examples and error fallback
• Collaborate with teams working on mental wellness and workplace empathy tools

Built With

• adk
• gmail-api
• google-calendar-api
• google-gemini-api
• notion-api
• prompt
• python