Chaitanya Asawa

Chaitanya “Chai” Asawa leads engineering for clinical decision support at Abridge, building systems that combine patient records, medical literature, and clinical conversations to help physicians make better-informed decisions. Previously a founding engineer at Glean, he helped develop the workplace AI assistant before applying his experience with context-dependent intelligence to healthcare.

Asawa began his career at Vicarious, working on robotics and AI research involving probabilistic graphical models. He joined Glean in 2019 and spent more than six years helping build its enterprise search product and Glean Assistant, eventually leading engineering teams responsible for the assistant. During his tenure, the company expanded from roughly ten employees to more than a thousand, a trajectory he described in his account of leaving Glean.

At Glean, he confronted a central challenge of workplace AI: useful answers depend on organizational context, including a person’s responsibilities, access permissions, colleagues, and scattered internal information. Healthcare presents a more consequential version of that problem. At Abridge, Asawa’s work connects electronic health records, live clinician-patient conversations, clinical guidelines, and medical literature to support chart preparation, clinical-trial matching, contextual medical questions, and clinician-approved orders.

• The clinical conversation anchors the workflow. Asawa treats the patient encounter as the context linking documentation, billing, care planning, and decision support. Systems need both longitudinal patient records and information emerging during the visit to provide relevant assistance before, during, and after appointments. His clinical-intelligence presentation illustrates this approach through voice-driven trial matching, chart preparation, and proposed medical orders.

• Clinician-directed intelligence preserves medical judgment. Asawa’s coauthored work on clinician-controlled documentation describes natural-language editing that lets physicians reshape generated notes around their reasoning and priorities. Automation can prepare drafts and recommendations, but clinicians retain responsibility for interpretation and approval.

• The generator-verifier gap complicates clinical evaluation. Assessing a nuanced medical answer can require nearly as much expertise as producing one. Asawa addresses this generator-verifier gap through physician-authored rubrics, independent clinical review, expert-calibrated model judges, safety and adversarial testing, staged releases, and continuous monitoring. Rubrics specify the elements a strong answer should contain without assuming every valid response must look identical.

• Specialized models balance quality, latency, and cost. Asawa decomposes documentation into narrower tasks, including individual note sections, and trains smaller models for specific workflows. Lightweight conversational triggers can determine when a larger system should match a proposed order against an approved catalog. His coauthored analysis of clinical foundation-model evaluation examines clinical accuracy, completeness, faithfulness, reasoning, and tool use together.

Asawa also advocates closer collaboration between engineers and practicing physicians. He helped organize a clinical-workflow AI hackathon involving Abridge, Anthropic, and Lightspeed, with clinician-builders, synthetic healthcare data, and model resources.