PharmaFair

Link for video doesnt go down there: https://www.loom.com/share/f083564ac8504835a8fef4ff103ae259

Inspiration

Sarah has epilepsy and depression. Her genome test flagged she was a poor metaboliser of SSRIs but that insight sat buried in a PDF, disconnected from her prescriptions. Her neurologist prescribed an antidepressant that interacted dangerously with her epilepsy medication. It took 6 weeks of side effects across 4 drugs before her team realised the combination was contraindicated by her own genetics. That story isn't rare over 100,000 Americans die from adverse drug reactions every year, most of them preventable. Genome tests exist. The data exists. But it's static, siloed, and useless at the point of care.

What it does

GenoStack connects a patient's genomic profile to their live medical situation. Rather than surfacing generic ancestry-style insights, it dynamically cross-references a user's current medications, conditions, and genetic markers to flag drug incompatibilities and recommend safer alternatives in real time. For Sarah, it would have flagged the epilepsy–antidepressant conflict on day one and surfaced a genetically compatible alternative no trial and error, no harm.

How we built it

We built GenoStack as a React Native mobile app with a Node.js backend. The core engine uses the Claude API to reason over a patient's genomic data, current medication list, and medical conditions translating complex pharmacogenomic interactions into plain-language recommendations. We used publicly available pharmacogenomics databases (PharmGKB) to ground the model's outputs in validated drug-gene interaction data.

Challenges we ran into

Making AI medical reasoning both accurate and appropriately cautious was the hardest design problem. We had to ensure GenoStack never oversteps, it flags risks and explains them, but always defers final decisions to a clinician. Structuring prompts that produced consistent, evidence-grounded outputs (rather than hallucinated drug advice) required significant iteration.

Accomplishments that we're proud of

Building a working prototype in under 3 hours that can take a simulated genomic profile + medication list and return a clinically-framed incompatibility report. The Sarah demo, walking through her exact scenario, produces a clear, actionable flag that a real patient could hand to their doctor.

What we learned

Pharmacogenomics is underleveraged not because the science isn't there, but because no one has built the bridge between genetic data and the clinical moment. The bottleneck is interface design and trust, not biology.

What's next for GenoStack

Integration with EHR systems so clinicians see genomic risk flags directly in the prescribing workflow. Expanding coverage beyond drug interactions to dosage optimisation based on metaboliser status. And a patient-facing version that empowers people to have informed conversations with their doctors — not replace them.