Vital Signal

AI agent that personalizes global health alerts based on your health, family locations, and travel plans, then sends a report along with translated family reports to protect loved ones across borders.

Comment 2

Inspiration

As an immigrant with family in Ethiopia and the Philippines, I understand the challenge of distance during health crises. When disease outbreaks happen, information exists but it's:

  • Broadcast to everyone (causing alert fatigue)
  • Only in English (excluding non-English speakers)
  • Generic (doesn't consider individual health risks)
  • Disconnected from family networks across countries

The core problem: 95% of people ignore global health alerts because they aren't relevant. The 5% who need urgent action miss critical warnings buried in the noise.

What it does

VitalSignal is an autonomous AI agent that personalizes global disease outbreak alerts using multi-factor risk analysis. It:

  • Analyzes 8+ personal factors per user (health conditions, family locations, travel plans, age, medications)
  • Makes autonomous decisions - Same alert → Different outcomes for different people
  • Generates medical intelligence - Converts ICD-10/SNOMED codes to patient-friendly explanations via PhenoML
  • Creates multilingual family reports - Translates comprehensive health guides to 20+ languages (Portuguese, Arabic, French, etc.)
  • Generates AI alert visuals - Freepik creates severity-color-coded medical posters
  • Sends personalized emails - Only alerts people actually at risk, with actionable guidance

Example: Dengue outbreak in Brazil → Maria (São Paulo, diabetic) gets MEDIUM alert in Portuguese. Sarah (NYC, pregnant, flying to Brazil) gets HIGH alert. John (Tokyo, healthy) gets nothing. True autonomous intelligence.

How I built it

Architecture: Airia multi-agent orchestration with 3 autonomous nodes

Node 1 - Data Collection:

  • Structify scrapes 189 real-time WHO/CDC health alerts
  • ClickHouse retrieves user profiles with family networks

Node 2 - Risk Analysis:

  • Custom FastAPI risk calculator with 8-factor scoring algorithm
  • PhenoML enriches with FHIR/SNOMED/ICD-10 medical intelligence

Node 3 - Smart Actions:

  • DeepL translates family health guides to user's family language
  • Freepik AI generates medical alert images
  • ClickHouse stores images as BLOBs
  • SendGrid delivers personalized emails with inline image attachments

Tech Stack:

  • Backend: FastAPI + Python 3.11
  • Database: ClickHouse Cloud (user profiles + image storage)
  • Orchestration: Airia (autonomous decision workflow)
  • APIs: Structify, PhenoML, DeepL, Freepik, SendGrid

Challenges I ran into

Challenge 1: Images not displaying in emails

  • Initial attempt: Base64 embedding (190KB) → Blocked by Gmail
  • Second attempt: ngrok public URLs → 405 errors
  • Solution: ClickHouse storage + SendGrid inline attachments with Content-ID references

Challenge 2: Achieving true autonomy

  • Problem: Most "AI agents" are glorified if-then workflows
  • Solution: Non-deterministic risk engine analyzing 8+ variables per user. Same alert produces different outcomes based on individual context - that's real autonomy.

Challenge 3: Useful family reports

  • Initial translations were just converted alerts - not actionable
  • Solution: Built comprehensive guides using PhenoML data: symptoms checklist, transmission info, prevention steps, emergency care guidance, family action plan - then translate entire guide.

Challenge 4: Localhost URLs in production emails

  • Problem: http://localhost:8000/images/123 works locally, fails externally
  • Solution: Store Freepik images in ClickHouse, retrieve at email-send time, attach as inline content

Accomplishments that I'm proud of

✅ Integrated 7 sponsor tools - Airia, Structify, ClickHouse, PhenoML, DeepL, Freepik, SendGrid working together autonomously

✅ True multi-agent autonomy - Non-deterministic risk decisions based on 8+ factors, not simple rules

✅ Real-world medical intelligence - PhenoML transforms medical jargon into patient-friendly explanations families can understand

✅ Cross-border family protection - Same user gets alerts in English, generates Portuguese report for sister in Brazil, French for family in Rwanda

✅ Production-ready architecture - ClickHouse image storage, async FastAPI, proper medical code standards (FHIR/SNOMED/ICD-10)

✅ 189 real health alerts - Structify scraping actual WHO/CDC data, not mock data

✅ AI-generated medical visuals - Freepik creates severity-color-coded alert images stored in database

What I learned

Technical:

  • Multi-agent orchestration requires careful state management between nodes
  • ClickHouse isn't just for time-series - excellent for BLOB storage and retrieval
  • Email inline attachments (Content-ID) solve image display issues better than external URLs
  • PhenoML's medical code enrichment is powerful for patient education
  • Autonomous decisions need 8+ contextual variables - anything less is just rules

Domain:

  • Risk is deeply personal - same outbreak affects people completely differently
  • Language barriers are a critical gap in global health alerts
  • Families are global networks that need cross-border health protection
  • Medical jargon (ICD-10 codes) means nothing to patients - plain language saves lives

Integration:

  • 7 different APIs require robust error handling and fallbacks
  • Image generation (Freepik) + storage (ClickHouse) + delivery (SendGrid) needs careful coordination
  • Translation context matters - "family" means different things in different cultures

What's next for Vital Signal

Behavioral Learning Engine

  • Adapt risk thresholds based on user feedback
  • Learn from false positives/negatives to improve accuracy

Proactive Travel Protection

  • Monitor flight bookings: "You're flying to Brazil next week - Dengue outbreak active, here's what to do"
  • Alert before travel, not during

Community Health Networks

  • Opt-in: If you're diagnosed, automatically alert your family/contacts
  • Create protective health networks across countries

FHIR Medical Record Integration

  • Pull actual health conditions from EHR systems
  • Real-time risk updates when medical status changes

Symptom Monitoring

  • Daily check-ins during active outbreaks
  • Early detection before severe illness

Multi-modal Alerts

  • SMS for urgent CRITICAL alerts
  • WhatsApp integration for family groups
  • Voice calls for elderly non-tech users

Please check demo here:https://drive.google.com/file/d/1YCWN7PK2r8o3UeF9A9W0dlnNLinGwTNH/view?usp=sharing

Use https://chat.airia.ai/agents/ for testing with email: gomezrichel99+1@gmail.com password: Hackathon123!

Built With

  • airia
  • clickhouse
  • deepl
  • fastapi
  • freepik
  • phenoml
  • sendgrid
  • structify
Share this project:

Updates