2023 - August 2024 · Co-founder · Healthcare AI · Past project
HEKADES
Hekim Karar Destek - AI-powered clinical decision support
HEKADES started with a conversation with professors from the University of Health Sciences (SBÜ). They told us that newly graduated doctors were increasingly turning to ChatGPT for clinical questions.
This was back when ChatGPT couldn’t even browse the internet. It could produce convincing answers, but there was often no reliable way to verify where the information came from, and hallucinations were a serious concern. That’s a particularly dangerous combination in healthcare.
We wanted to bring structure and verifiability to AI-generated medical answers.
HEKADES was designed to research trusted medical literature based on a patient’s clinical situation, retrieve relevant information, and provide doctors with answers grounded in identifiable sources.
Building HEKADES
We were working at a time when retrieval-augmented generation (RAG) was rapidly evolving. A considerable part of our work involved following new research papers, experimenting with retrieval algorithms, and figuring out how to apply those ideas to clinical literature.
The challenge wasn’t simply getting an LLM to produce a plausible answer. It was making the underlying research process more structured, traceable, and useful to medical professionals.
We developed HEKADES with input from medical professionals through KWORKS, Koç University’s entrepreneurship center, and presented it to healthcare organizations and potential investors.
Ultimately, licensing restrictions around medical guidelines prevented us from bringing the product to market.
HEKADES was my first experience working so closely with emerging AI research while confronting the practical constraints of a high-stakes domain.