HealthShield - AI Tinkerers - Kuala Lumpur Hackathon
AI Tinkerers - Kuala Lumpur
Hackathon Showcase

HealthShield

Team consisting of AI Engineer (CelcomDigi), Monash AI master's student, and Software Engineer (Dassault) — expertise: LLMs, PyTorch/TensorFlow, React, Spring Boot, Python.

3 members

Track: 3 – The Proactive Public Health Shield

Problem: Public health officials in Malaysia often detect outbreaks too late due to slow, manual analysis of scattered news and social media. This delays response and risks public safety.
Our Solution: A real-time AI-powered dashboard that detects trending health threats, evaluates risk, summarizes context, and suggests actions that turns reactive responses into proactive protection.

Target User: Ministry of Health analysts, Public health decision-makers (ministers, epidemiologists, hospital administrators)

Workflow

  • Data Ingestion: Live scraping of Malaysian news feeds i.e Top Malaysia news channel Malaymail, malaysiannews, etc and social media post (future works: ASEAN news)
  • Keyword Extraction: AI detects trending health-related topics
  • Risk Evaluation: AI assigns threat levels (per topic & overall)
  • Insights: AI summaries + recommended actions
  • Visuals: Trends chart & state-level heatmap

Key Outcomes

  • Time Saved: Days of manual analysis reduced to minutes
  • Faster Response: Detect threats before escalation
  • Scalable: Extend to ASEAN, multi-language support

Judging Criteria

  • Innovation: First-of-its-kind AI dashboard for Malaysian public health threat monitoring
  • Usability: User friendly interface with visualization and interactive UI
  • Feasibility: Functional prototype with live Malaysian data
  • Technical Merit: Groq AI + real-time scraping + visual analytics
  • User Benefit: Clear, actionable insights, time saving
  • Deployability: Modular, cloud-hosted, expandable, ready to be deployed on production

Tech Stack
Python (FastAPI), Groq AI (summarization, evaluation, calculation), Next.JS, ShadCN, Leaflet.js, HuggingFace (sentence transformer for document correlations), Docker.
GroqAI model <- “llama-3.3-70b-versatile” with 12k max token as we need to feed as much data as possible
HuggingFace <- “all-MiniLM-L6-v2” this model get articles that are related semantically to the keyword

Quick Demo Steps

  1. Open dashboard (URL)
  2. See today’s trending keywords + risk scores
  3. Click keyword → AI summary, trends, recommended actions
  4. Filter by state → localized impact map