In exciting news from Paris, the startup Nabla has just announced its collaboration with CMA CGM’s corporate VC fund – ZEBOX Ventures, leading a $24 million Series B funding round with Cathay Innovation. This funding comes after months of negotiation for a significant partnership with the prominent medical group, Kaiser Permanente, a division of the leading American healthcare company dedicated to health management.
According to sources, Nabla’s current funding round has propelled its valuation to an impressive $180 million. The company has the potential to gather additional funds from American investors during this phase.
Nabla is making waves in the medical field by developing an AI copilot for doctors and other medical professionals. The best way to describe it is as a silent work partner sitting in the corner of the room, taking notes, and crafting medical reports for you.
Founded by Alexandre Lebrun, Dolphin Groult, and Martin Raïs, Lebrun, the CEO of Nabla, was previously the CEO of Wit.ai, an AI assistant startup acquired by Facebook. Afterward, he became the head of engineering at Facebook’s AI research lab, FAIR.
Just a few weeks ago, I had the opportunity to witness a live demo of Nabla in action. In the demo, a real doctor and a simulated patient pretended to have lower back pain. As soon as the consultation began, Its’s interface was activated with the press of a button, and the doctors seamlessly forgot about their computers.
Beyond physical examinations, the consultation involves a detailed discussion about your medical history with a group of questions. At the end of the session, prescriptions and follow-up appointments can be arranged.
Nabla employs speech-to-text technology to transcribe conversations, facilitating both telehealth appointments and personal consultations. After the patient leaves, the doctor presses a stop button, and Nabla leverages advanced language models to enhance medical discussions and details related to health.
Nabla Copilot strives not to replace human involvement in medical processes but rather aims to assist doctors in saving time on administrative tasks, allowing them to focus more on patient care.
Lebrun emphasized, “We don’t want to change doctors in the near future. You’ve seen companies like Babylon in the UK trying chatbots and attempting immediate automation of things, taking loops away from doctors. And we decided a long time ago with Nabla Copilot that [doctors] are in the loop, and we work shoulder to shoulder with them.”
He went on to say, “It’s somewhat akin to self-driving cars operating at level two. We’re still at level one today. We’re going to start with clinical insurance support very soon. Then level three is clinical decision support, but with FDA approval, as you make decisions that you can’t really explain.”
At some point, one might imagine the concept of autonomous healthcare, meaning doctors being removed from the room. However, Lebrun remains cautious on this front. He expressed, “For certain markets and situations, such as in countries without healthcare access, it could be a pertinent consideration.” Ultimately, he sees diagnostic work as a “pattern matching problem” that can be solved with AI, allowing doctors to focus on empathy, surgical procedures, and crucial decisions.
While Nabla is based in France, most of its clientele is in the United States, especially after rolling out with the prominent medical group. It is not just another tool; it’s actively used by thousands of doctors every day.
Nabla’s Confidentiality Model
Nabla is currently available as a web app or Google Chrome extension, designed with a keen awareness of handling sensitive data. The company takes extra measures to ensure that it does not store audio or medical notes on its servers unless both the doctor and the patient provide consent.
Rather than focusing on data storage, Nabla emphasizes data processing. After a consultation, audio files are immediately deleted, and the transcript is securely saved in the Electronic Health Record (EHR) for the doctor’s reference.
In technical terms, when a physician initiates recording, the audio is instantly transcribed in real-time using the Speech-to-Text API. Nabla utilizes a combination of Microsoft Azure’s Speech-to-Text API and its own Speech-to-Text model (built on the open-source Vosk model).
“Having just a generic Speech-to-Text algorithm may not be accurate on medical data. But we have fine-tuned ours. As you might have noticed, the text starts off very light and then darkens. When it darkens, it means we’ve validated it against our model and aligned it with drug names or medical conditions,” explained Nabla ML Engineer Gregoire Retorn during the demo I witnessed.
The transcript is initially anonymized, meaning personally identifiable information is replaced with variables. Actions are taken on these anonymized transcripts using a large language model. Historically, Nabla has used GPT-3 and then GPT-4 as its primary large language models. As an enterprise customer, Nabla can inform OpenAI that it cannot store its data and cannot train its large language model on these consultations.
However, Nabla is also playing with the idea of using more restrictive models against generic models in the future, as Lebron mentioned, “In the future, we envision using more constrained models against generic models.”
Once the Large Language Model (LLM) has performed actions on the transcript, Nabla separates the output from the anonymization. Doctors can view the note, which is stored locally in the web browser’s storage files. Notes can then be imported into EHRs.
Nevertheless, doctors can provide consent and request patient satisfaction for sharing medical notes with Nabla to rectify transcription errors. With Nabla actively handling over 3 million consultations in more than 30 languages each year, the potential for Nabla to continually improve in the real world is quite promising.
For More Info: SumUp Secures €285M
