Janie Lee

Janie Lee is Abridge’s vice president of product, working on AI products built around patient–clinician conversations. Abridge announced her appointment on January 13, 2026. In a May 2026 interview, she described the company’s expansion from reducing documentation work toward helping health systems improve finances and patient outcomes, with clinical decision support being built by her colleague Chaitanya “Chai” Asawa.[3][1]

Lee argues that healthcare AI should help clinicians concentrate on patients. Her analogy is “we want our product to feel like air conditioning”: useful in the background, intervening when timing and clinical risk justify it. Preparing a clinician with relevant patient context before a visit can avoid repeated interruptions during a sensitive conversation. She also said patients and clinicians did not want a literal third voice in the room.[1]

She illustrated that approach with a hypothetical patient prescribed an MRI for knee pain. Rather than discovering missing authorization requirements weeks later, the product could identify questions about prior physical therapy and the duration of pain while the patient was still present. Existing context would supply some answers; the clinician could collect the remaining information before the patient left. Lee presented this as a product opportunity to bring delayed administrative work into the visit.[1]

Lee explained that such assistance requires electronic health record (EHR) context, including previous laboratory results and imaging, alongside payer policies that vary by state and may reside in unstructured, 50-page PDFs. Reliable model output is only part of the problem: workflow timing determines whether the system can resolve missing information when the patient is available. She similarly treats EHR interoperability as essential to adoption, describing close partnerships that enabled data exchange through APIs that were not ready out of the box. Extra clicks, she said, could be enough to discourage clinician use.[1]

Lee describes personalization at three levels: the individual clinician, the specialty, and the health system. Individual preferences include note length, bullets versus paragraphs, phrases, and templates, but those preferences must preserve accuracy. Specialty differences reach beyond presentation: a complete, compliant, billable dermatology note requires different evaluation criteria from a primary-care note. Health systems also want their own clinical guidelines incorporated into decision support. Lee identifies clinicians’ edits as feedback that can deepen personalization over time.[1]

Her account of Abridge’s evaluation process begins with defining good output when a feature is first developed. In-house clinicians inspect initial outputs through an internal process called LFD—“look at the effing data.” LLM judges are then calibrated using annotated data and internal or external evaluators, with additional human evaluation according to the stakes. The relevant expertise varies by task: products intended to improve billable documentation may call for healthcare coding teams rather than clinicians. Lee emphasizes operational planning and evaluator capacity alongside the machine-learning work.[1]

Lee credits embedded clinician scientists—clinically trained colleagues with technical skills—with assessing clinical usefulness and helping define evaluation criteria. She explicitly distinguishes their expertise from her own and Asawa’s: neither has a clinical background. She also describes the work of deciding when hundreds or thousands of offline responses are needed to catch long-tail cases before production.[1]

In May 2026, Lee said Abridge had moved customers to monthly release cycles, with a subset willing to develop and provide feedback outside those cycles. She contrasted the continuing high bar for healthcare releases with the frequent experiments she had run at Loom. Healthcare also changed her expectation about AI progress: she initially thought the field would adopt innovations late, but came to believe that its low tolerance for errors would force solutions to some difficult AI problems earlier.[1]

Lee reconsidered the idea that prototypes could replace product requirements documents after her team tried changing its development approach. Prototypes did not adequately capture data-use constraints, customer implementation requirements, or the strategic question of whether a problem deserved the company’s time. Her conclusion was that “crisp written clarity is more important than ever”, even if that writing now lives in a Markdown file that teams and systems can use as context.[1]

She still values prototypes for exploring alternatives and obtaining clinician feedback. Written documentation serves a complementary purpose: making implementation details, security, compliance, and edge cases explicit before a product reaches customers. Lee argues that AI should help create that documentation, while product teams retain responsibility for deciding why a problem matters and how solving it differentiates the company. She described the gap between a weekend demo and a product that works at major health systems as substantial, while saying she was building faster than at any earlier point in her career.[1]

Before Abridge, Lee worked on product leadership, identity management, hardware, and real-estate pricing. She drew on those earlier experiences in the healthcare interview, contrasting Loom’s experimentation with clinical rollout requirements and recalling how pricing-model outliers at Opendoor could consume margins.[4][1]

At Opendoor, Lee launched markets and developed pricing algorithms. She subsequently led Rippling’s Identity Management and Hardware teams. Atlassian’s Team ’24 Europe biography also lists earlier enterprise and consumer product experience at Box.[4][5]

By June 2024, Lee led product at Loom and owned its Self-Serve business, according to her 20Product biography. At Team ’24 Europe, she was listed for the session “Unleash every team’s potential with Loom AI.”[4][5]