Noa Notes is one of the most interesting projects I work on. It’s a medical assistant that listens to conversations between doctors and patients and generates an accurate, professional, and useful summary of the appointment, which the doctor can immediately use as a record of the visit.
This project has everything: summarization in a challenging setting, medical transcription, fine-tuned open-source models, advanced evaluation, complex ground truth datasets - and more. Any one of these aspects would be a great starting point for a solid half-hour talk. A few months ago, I suggested to Docplanner that we share some of our findings, challenges, and solutions with a wider audience. I eventually joined forces with Karol Traczowski, Head of AI at Docplanner, and we submitted our presentation for the 2024 AI Summit conference in Warsaw.
I am happy and proud that our talk was recognized as one of the most interesting presentations of the event. To be precise, we were ranked as the second most interesting session - right behind Cezary Kuik’s lecture on enhancing existing products and services with LLMs, and ahead of the “Deep Networks in Waterworks Infrastructure” presentation by Władysław Dzik. There were actually many other insightful presentations, which makes this recognition even more meaningful and rewarding.
Unfortunately, the session was not recorded. During our 20 minutes, we covered a lot of ground: trends in healthcare, the value proposition the tool offers to healthcare professionals, technical architecture, challenges, growth metrics, and even testimonials from doctors. It was quite difficult to squeeze everything into such a short time, so we had to prepare diligently, time each other, and cut out any fluff. Quite recently, Karol also did a podcast about Noa (which wasn’t called Noa yet :)) where he covered some of these topics in more depth (the conversation is in Polish).