AI-Health-Companion

AI Health Companion with a Fivetran custom connector that streams anonymized Q&A telemetry and curated health content into BigQuery . He use Vertex AI/Gemini for LLM answers and augmented analytics.

Comment

AI Health Companion

Inspiration

Our team wanted to build something truly helpful in everyday life — a trusted AI companion for health-related questions, powered by real data, not just language models.
We noticed that people often turn to the internet for health advice but rarely get answers backed by reliable information.
That inspired us to combine Google Cloud’s Vertex AI (Gemini) with Fivetran’s automated data pipelines and BigQuery analytics, to deliver not only answers, but also insights.

What it does

AI Health Companion is a conversational assistant that:

  • Lets users ask health or lifestyle questions in natural language.
  • Generates accurate, cited answers using Gemini on Vertex AI.
  • Stores anonymized interaction data in BigQuery via a custom Fivetran Connector.
  • Visualizes daily usage and top health topics in an Angular dashboard.
  • Continuously ingests new verified health content daily through an agentic workflow using Vertex AI summarization.

How we built it

Tech Stack Overview

  • Frontend: Angular 18 (standalone components, Signals, Data Viz) → deployed on Netlify.
  • Backend: Node.js/Express on Cloud Run → handles /api/ask, /etl, /analytics/snapshot.
  • Connector: Custom Python service (FastAPI) built with Fivetran Connector SDK, deployed to Cloud Run.
  • Data Layer: Fivetran → BigQuery dataset (ai_health) → views for analytics.
  • AI Layer: Vertex AI (Gemini 1.5 Flash) for real-time Q&A, summarization, and auto-tagging of new content.
  • Automation: Cloud Scheduler triggers a daily ingest job → Gemini annotates articles → Fivetran syncs data into BigQuery.
  • CI/CD: GitHub Actions deploys backend & connector to Cloud Run, and frontend to Netlify.

Challenges we ran into

  • Setting up the Fivetran custom connector with correct pagination and schema discovery logic.
  • Handling Vertex AI authentication on Cloud Run without exposing keys (using service accounts).
  • Ensuring CORS and proxy rules worked seamlessly between Netlify and Cloud Run.
  • Managing schema consistency in BigQuery when syncing incremental data.
  • Designing a lightweight but meaningful analytics dashboard that works both locally and in production.

Accomplishments that we're proud of

  • Built a fully working end-to-end AI data pipeline in less than two weeks.
  • Integrated Fivetran → BigQuery → Vertex AI smoothly for continuous data enrichment.
  • Created a clean Angular UI with real-time insights (usage trends, feedback, sources).
  • Deployed everything serverlessly — fully running on Google Cloud Run and Netlify.
  • Implemented a daily agentic ingestion flow for health content using Gemini.

What we learned

  • How to design LLM-driven apps that combine conversational AI with structured analytics.
  • The power of Fivetran SDK for creating new connectors and streaming telemetry.
  • Best practices for secure AI deployments with service accounts and Secret Manager.
  • How to build maintainable CI/CD pipelines with GitHub Actions → Cloud Run → Netlify.
  • That small, focused teams can achieve production-ready integrations surprisingly fast!

What's next for AI Health Companion

  • Expand the knowledge base with trusted medical sources and research articles.
  • Implement personalized dashboards (e.g., by topic or user interest).
  • Add multi-language support for broader accessibility.
  • Integrate Elastic hybrid search to ground Gemini’s answers with richer context.
  • Build a mobile companion app with offline summaries and push notifications.

Built With

Share this project:

Updates