Inspiration
When we first looked into Thalassemia, we expected to find a lack of access to blood or infrastructure. What we didn’t expect was something much deeper — the quiet, invisible cycle of inherited pain. Families discovering only after childbirth that both parents were carriers. The helplessness of knowing it could’ve been prevented, if only they had known.
That hit hard.
We didn’t want to build another donor tracker. We wanted to intervene before pain begins, before there’s a hospital room, before there’s a diagnosis. What if technology could help people avoid a genetic tragedy they didn’t know they were walking into?
That question became our mission.
What it does
GeneMatch AI is not your typical healthcare app. It’s a preventive AI system that works quietly in the background to do something powerful:
Help people make informed choices before Thalassemia becomes a part of their child’s future.
Here’s what it does:
Predicts carrier likelihood based on anonymized family and health data — no blood test needed.
Forecasts donor availability using behavioral models (because donor burnout is real).
Identifies high-risk communities and suggests when and where to run screening drives — based on local festivals, marriage seasons, and health trends.
Uses LLMs to create hyper-local awareness messages that feel like they’re coming from someone you know, not a hospital.
How we built it
We had a short time and a big dream — so we split it into four focused builds:
Predictive Carrier Score: We built a model trained on health/demographic traits that could provide a probability of being a carrier. It’s not a diagnosis — just a nudge to take action.
Donor Availability Engine: Using LSTM and seasonal data, we built a system that doesn’t just track donors — it predicts who’s likely to donate again and when.
Outreach Copilot: We trained OpenAI to speak like a local — generating real, relatable messages in local languages that can be dropped into WhatsApp communities.
Heatmap Intelligence: We used Azure Maps to overlay high-risk zones with behavioral data and suggested smart windows for running awareness or testing campaigns.
Challenges we ran into
We had to be careful — we're dealing with genetics, community trust, and sensitive outreach. Some challenges that tested us:
Lack of real datasets — we didn’t want to fake it, so we worked with simulated + research-based patterns.
Avoiding fear-based messaging — our AI had to be empathetic, not alarming.
Balancing ambition and reality — not everyone has a smartphone, not everyone trusts tech. We had to think about low-tech bridges.
Accomplishments we’re proud of
We built something that’s not flashy, but quietly powerful.
We didn’t just think about the problem — we questioned why the problem keeps repeating
How we built it
We had a short time and a big dream — so we split it into four focused builds:
Predictive Carrier Score: We built a model trained on health/demographic traits that could provide a probability of being a carrier. It’s not a diagnosis — just a nudge to take action.
Donor Availability Engine: Using LSTM and seasonal data, we built a system that doesn’t just track donors — it predicts who’s likely to donate again and when.
Outreach Copilot: We trained OpenAI to speak like a local — generating real, relatable messages in local languages that can be dropped into WhatsApp communities.
Heatmap Intelligence: We used Azure Maps to overlay high-risk zones with behavioral data and suggested smart windows for running awareness or testing campaigns.
Challenges we ran into
We had to be careful — we're dealing with genetics, community trust, and sensitive outreach. Some challenges that tested us:
Lack of real datasets — we didn’t want to fake it, so we worked with simulated + research-based patterns.
Avoiding fear-based messaging — our AI had to be empathetic, not alarming.
Balancing ambition and reality — not everyone has a smartphone, not everyone trusts tech. We had to think about low-tech bridges.
Accomplishments that we're proud of
We built something that’s not flashy, but quietly powerful.
We didn’t just think about the problem — we questioned why the problem keeps repeating, and tackled that.
Our idea doesn’t stop with one region, one group, or one year. It scales across communities and generations.
What we learned
Technology isn’t useful unless it understands people.
Prevention doesn’t get the headlines — but it changes lives in silence.
Sometimes, the right solution isn’t the most obvious one — it’s the one that whispers instead of shouts.
What's next for GeneMatch AI
Partner with Blood Warriors to test the outreach engine in one real-world community.
Build a microapp for health workers — no login, no installs — just a simple interface they can trust.
Train GeneMatch with actual anonymized genetic datasets once available.
Build offline-first messaging tools to reach regions without strong internet.
Expand beyond Thalassemia — to sickle cell, other inheritable disorders — where this system can quietly save futures.
Built With
- azure
- cosmos
- css3
- db
- e-raktkosh-api-(planned)
- firebase-firestore
- html5
- hugging-face
- javascript
- lstm
- maps
- openai-api
- plotly
- power-bi
- python
- pytorch
- scikit-learn
- transformers
- whatsapp-cloud-api
Log in or sign up for Devpost to join the conversation.