CompassionateConnect AI

• 

  Compassionateconnect AI - AI supports care, not replaces it.

• 

  This is the intake_form

Inspiration

As someone seeking therapy, I was overwhelmed by paperwork, repeated questions, and intake forms that felt impersonal. I often found myself having to explain the same concerns to multiple people. I wondered: could AI reduce that friction — and actually make the first step feel more human? CompassionateConnect AI was born from that question.

What it does

CompassionateConnect AI is a Gemini-powered, multi-agent intake system for mental health clinics. It:

• Asks patients for their name, mood, and concerns

• Clarifies unclear answers using Gemini

• Detects signs of crisis and flags them for staff

• Summarizes responses in plain language

• Offers ethical, non-diagnostic insights to therapists

• Stores data securely in both Firestore and JSON for therapist dashboards

The system runs fully in a command-line interface, designed for clinics without web developers or complex infrastructure.

How I built it

I used Google Vertex AI’s Gemini 1.5 Flash model to build multiple agents in Python:

• An IntakeAgent for asking questions and handling skipped inputs

• A CrisisResponseAgent that detects risk language and logs high-priority patients

• A SummaryGeneratorAgent that creates readable summaries of each intake

• An InsightAgent that provides possible therapy approaches

• A DataPersistenceAgent that stores everything in Firestore and JSON

I used Google Cloud SDK, Firestore, and local files to keep it lightweight and deployable from any terminal.

Challenges I ran into

• Debugging Gemini API errors and authentication scopes

• Designing a CLI-based multi-agent system with ethical guardrails

• Making the Firestore sync bidirectional and safe

• Writing a summary pipeline that was actually helpful without overstepping into diagnosis

Accomplishments that I am proud of

• Fully functional multi-agent pipeline with Gemini 1.5 Flash

• Working CLI interface for intake and therapist dashboard

• Crisis detection and ethical insight generation

• Clean architecture, clean voiceover demo, and Notion documentation

• Personalized — and born from real experience

IMPORTANT:

Currently, CompassionateConnect AI runs on a Render platform for the onboarding_process intake form and the therapist_dashboard primarily via a command-line interface (CLI) that guides you through the intake process. T

What I learned

• How to design AI systems with real human stakes in mind

• How to use Vertex AI + Firestore + Python to build fast prototypes

• The importance of data clarity and trust when working in mental health

• That a CLI can still feel compassionate if designed with care

What's next for CompassionateConnect AI

• Add a lightweight web-based interface (for clinics without technical staff)

• Enable follow-up agent automation and scheduling integrations

• Train Gemini prompts on clinic-specific language

• Expand insight generation into client preparation tools

• Partner with therapists or nonprofits for real-world pilots

Built With

• adkhackathon
• firestore
• google
• google-cloud-sdk
• google-vertex-ai-(gemini-1.5-flash)
• googleadk
• json
• python