#58, Using AI to Transform Medical Documentation, DeepScribe's Akilesh Bapu
Aug 9, 2023
auto_awesome
Akilesh Bapu, Co-Founder & CEO of DeepScribe, discusses transforming medical documentation with AI. DeepScribe uses supervised and unsupervised machine learning techniques to enable physicians to spend more time with patients. Their go-to-market strategy targets private practices and large health systems. Maintaining a 'human in the loop' is crucial to accuracy and managing algorithmic bias in AI systems.
DeepScribe uses AI to transform medical documentation, enabling physicians to spend more time with patients instead of manual note-taking.
DeepScribe emphasizes the importance of maintaining a 'human in the loop' to ensure accuracy and manage algorithmic bias in AI systems.
Deep dives
DeepScribe: Using AI to Transform Medical Documentation
DeepScribe, a company using AI to transform medical documentation, aims to solve the problem of time-consuming and manual note-taking for healthcare professionals. By leveraging supervised and unsupervised machine learning techniques, DeepScribe enables physicians to spend more time with patients and less time on documentation. The company focuses on an inbound-heavy go-to-market strategy, initially targeting private practices before expanding to large health systems. DeepScribe believes in maintaining a human in the loop in AI systems to ensure accuracy and manage algorithmic bias. With the goal of becoming a data company of healthcare, DeepScribe aims to collect a foundational dataset that will unlock various healthcare solutions beyond documentation.
DeepScribe's Approach to AI and Technology
DeepScribe employs a unique approach to AI by transcribing audio using multiple vendors to ensure high accuracy. They use a combination of large language models and their dataset, such as GPD4, to generate medical notes. Before the notes are delivered to clinicians, they undergo review by a human to bridge the gap between AI's production readiness and optimal accuracy. DeepScribe believes in both supervised and unsupervised machine learning for healthcare tasks, combining labeled data to guide models while leveraging unsupervised learning for broad language understanding. Their platform provides clinicians with flexibility, reduces documentation time, and aims to enhance patient care.
The Role of DeepScribe's Human Loop in AI Training
DeepScribe's human loop, consisting of medically knowledgeable scribes, plays a vital role in optimizing AI training and model accuracy. These scribes not only write notes but also label the data. DeepScribe sources talent from pre-medical students, both locally (such as Berkeley) and internationally (India, Philippines), ensuring high-quality data labeling. They prioritize a product-focused approach to improve the human loop experience, which is instrumental in producing high-quality data that leads to better AI. Scaling the human loop effectively requires starting with a scrappy approach, validating business fundamentals, and gradually investing in quality standards and scalability.
Ensuring Accuracy, Bias Mitigation, and Customer Delight
DeepScribe prioritizes accuracy and bias mitigation in its AI systems. Through continuous reinforcement learning and using the human loop as a safeguard, they address biases within their models. DeepScribe acknowledges the challenge of eliminating bias completely but emphasizes the importance of data cleanliness and human control to minimize biases. Concerning customer delight, DeepScribe recognizes the crucial role of delivering a product that truly solves customer pain points and enhances the overall healthcare experience. They are committed to designing products that align with end-user needs and prioritize patient care, aiming to avoid repeating the mistakes of early EHR implementations and delivering a seamless workflow for clinicians.
In this episode, host Alex Merwin, Head of Growth for Healthcare & Life Science Startups, welcomes Akilesh Bapu, the Co-Founder & CEO of DeepScribe, an AI medical scribe company revolutionizing medical documentation. DeepScribe, trusted by both large and small healthcare systems across America, brings passion and determination to provide the most trustworthy AI medical scribe available, aiming to bring the joy of care back to medicine.
We discuss the details of DeepScribe's AI-based approach to transform medical documentation. From using both supervised and unsupervised machine learning techniques and large language models, DeepScribe is focused on enabling physicians to spend more time with their patients rather than on manual note-taking.
Further, we explore DeepScribe's go-to-market strategy. Starting with an inbound heavy approach targeting private practices, DeepScribe then expanded to sell into large health systems. This approach has helped the company position itself as a leading AI medical scribe.
Finally, we discuss one of the most critical aspects of AI application in healthcare settings: maintaining a 'human in the loop'. Ensuring the accuracy of AI systems and managing algorithmic bias is critical, and DeepScribe has managed to integrate this principle into their business and product.
[01:44] Akileh's background as an AI researcher and what brought him to founding DeepScribe.
[04:46] How DeepScribe thinks about AI and how it's solutions works.
[08:40] Physician burnout and why we need more efficient tooling, and the importance of having humans in the loop to accurately ground truth data.
[14:41] How DeepScribe used an SEM/SEO driven inbound GTM strategy to grow market segment share with private practice, later expanding into health systems.
[17:22] How giving time back to physicians improves their work life harmony, benefitting them, their patients and the health system more broadly.
[19:41] DeepScribe's approach to managing the tradeoffs between individualizing care and perpetuating systemic bias which can be a challenge for AI systems.
[22:47] How DeepScribe leverages both supervised & unsupervised learning, and how "scribes" help them ground truth data at scale.
[28:27] How DeepScribe approached its different rounds of financing, from pre-seed to Series B.