HemaLinkAI

HemaLinkAI predicts donor availability, matches them to Thalassemia patients in real-time, and boosts recurring donations using AI, LLMs, and emotional storytelling.

Comment

Inspiration

Thalassemia patients often need blood every 15–20 days throughout their lives. In India, many face last-minute unavailability of donors, poor tracking systems, and low awareness. While initiatives like Blood Warriors are making a difference, their processes are still manual and reactive.
We wanted to build a technology that could predict, automate, and emotionally engage — not just connect patients to donors, but anticipate needs and strengthen the donor ecosystem.

What it does

HemaLinkAI is a smart AI-powered platform that:

  • Predicts when specific patients will need blood based on past transfusion intervals
  • Identifies and ranks potential donors using time-series and graph-based models
  • Matches donors and patients in real-time using location, blood group, and history
  • Sends personalized emotional prompts to donors using LLMs
  • Integrates with existing systems like e-RaktKosh and Blood Bridge for streamlined operations

How we built it

  • Frontend: React.js interface for patients and donors
  • Backend: Python (FastAPI), Supabase for database and auth
  • AI Modules:
    • Donor Prediction: Time-series model using Prophet
    • Matching Engine: Graph Neural Networks with PyTorch Geometric
    • LLM Message Generator: LangChain + RAG to craft personalized emotional messages
  • Data Pipeline: Simulated donor and transfusion datasets, cleaned and preprocessed in Python
  • Integration: API stubs to connect with e-RaktKosh and Blood Warriors systems

Challenges we ran into

  • Lack of real public donor datasets — we had to simulate data and validate models synthetically
  • Graph models required careful feature engineering to ensure effective patient–donor links
  • Building emotional messages using LLMs while maintaining empathy and medical professionalism
  • Integrating AI predictions with a live matching workflow without manual overrides

Accomplishments that we're proud of

  • Built a fully working end-to-end AI system in under 48 hours
  • Created predictive models that can anticipate donor unavailability with over 80% accuracy
  • Successfully matched patients with top 3 donors in under 5 seconds
  • Integrated LLMs to generate sensitive, personalized donor prompts
  • Designed a scalable system that could plug into existing national blood donation platforms

What we learned

  • Real-world healthtech systems require not just performance but empathy
  • Graph-based models are powerful when connections matter more than data volume
  • Prompt engineering plays a huge role in crafting human-centric AI solutions
  • Predictive intelligence is only impactful when paired with actionable interfaces
  • Coordination between AI and human workflows is critical in life-impacting domains

What's next for HemaLinkAI

  • Partner with Blood Warriors to test the platform with real users
  • Integrate directly with e-RaktKosh and state-level health APIs
  • Add WhatsApp/SMS-based nudging for donors who are not on the app
  • Expand models to cover other conditions like Sickle Cell Anemia
  • Deploy LLMs with feedback loops to ensure emotionally responsible messages
  • Open source the core modules to encourage nationwide adoption
Share this project:

Updates