RareSum: mining years of clinical notes to catch rare diseases earlier
A Turkiye-based startup runs NLP over messy hospital notes to surface rare-disease clues that no single doctor could ever read at scale.
I ran into RareSum in a list of finalists for the UN's AI for Good Innovation Factory, and the number that stopped me had nothing to do with the technology: patients with a rare disease wait five to seven years, on average, for an accurate diagnosis, and nearly four in ten get misdiagnosed at least once along the way. Multiply that by the roughly 350 million people living with a rare disease worldwide, and diagnosis time itself becomes the product to fix, not a new drug.
RareSum, a health-tech startup based in Turkiye, is building that diagnosis layer. Its platform pulls both structured and unstructured records from a hospital, then runs the unstructured half (the free-text notes doctors actually write) through medical-grade NLP and fine-tuned transformer models with ontology mapping, converting messy prose into standardized phenotypes a system can reason over. Those signals feed what the company calls its rare disease algorithm library, which returns risk scores, differential diagnoses, and case-based reasoning a clinician can read and challenge, not a black-box verdict.
The clever part is where the missing information actually lives. It's rarely absent, it's just buried: a symptom pattern one specialist wrote down three years earlier, in a hospital across the country, in a format nobody's software was built to read at scale. RareSum's bet is that recognition, not new medicine, is the bottleneck, and that a clinician paired with a system that has effectively read every note ever written can close a gap no single doctor can close alone.
Turkiye isn't a random home base either: higher rates of consanguineous marriage there push up the prevalence of certain rare conditions, giving the team a dense local testbed before expanding into Europe and, from 2027, MENA, the US, APAC and Latin America.
Investors reportedly took eighteen months to warm to the pitch, tripped up by the word "rare" itself, until someone pointed out that 350 million people is not a niche market. That's the whole story in one sentence.