Surgical Video Summarization Service

The summary was only useful if the surrounding service worked

The interface could not solve who recorded the procedure, how the source moved securely, what evidence the summary preserved, or who remained responsible for the decision. Those dependencies shaped the product as much as the summarization model did.

Connect the source to review instead of replacing it

The future-state service kept the original recording available and made the proposed summary reviewable in context.

1. Capture
2. Secure transfer
3. Summarize
4. Review against source
5. Store or share
6. Clinician decides

The system reduced the search space; the clinician owned the judgment

Residents provided source context. AI proposed relevant segments. Service infrastructure protected access and recovery. Attending clinicians reviewed the evidence and retained responsibility.

Map the clinical workflow before defining the interface

The Chronicle team combined secondary research, contextual inquiry, and interviews with ENT surgeons and residents. The work connected observed friction to specific service touchpoints rather than treating summarization as an isolated feature.

Translate service requirements into an inspectable interaction model

Thirteen priority interactions created a shared design target. Wireframes and successive iterations then tested how navigation, playback, annotation, and evidence states could work together.

Test the interaction flow, then make evidence states accessible

Evaluative testing with peers and the sponsor informed the final interaction flow. The delivered concept distinguished AI-suggested, manually created, and bookmarked clips and documented accessibility considerations.

Test whether the workflow could survive institutional reality

After Chronicle, I co-founded Brevity AI and served as CPO and sole UX lead. I worked with a part-time CTO and clinical, technical, legal, compliance, IRB, and incubator stakeholders to translate the concept into a minimum flow across capture or retrieval, secure upload, summarization, review, and storage.

1. User value
2. Technical feasibility
3. Workflow readiness
4. Reviewer confidence
5. Institutional fit
6. Adoption risk

Consultation support advanced because it fit roles and incentives

Procedure documentation was deprioritized, conference preparation remained exploratory, and malpractice-risk reduction failed validation. Consultation support advanced because it fit existing work more closely.

Pilot confidence was encouraging, but bounded

The public record describes six ENT consultation videos and reports unanimous 5/5 confidence in the summarized-video condition. Exact reviewer count, note scores, and the rating instrument were not preserved, so this remains a promising confidence signal rather than evidence of clinical efficacy.

Separate what was completed, tested, proposed, and not claimed

The work established the service model, high-fidelity interaction patterns, an operational MVP flow, use-case decisions, and a bounded pilot signal. Brevity closed because of funding constraints before broader institutional deployment could be tested.

1. Completed: research synthesis, service mapping, high-fidelity flows, and an operational MVP workflow
2. Tested: a surgeon-reviewed concept, use-case hypotheses, and a six-video consultation pilot
3. Proposed: failure behavior, confidence signals, and storage optimization
4. Not claimed: clinical efficacy, regulatory approval, measured time savings, scaled adoption, or EHR integration

Place AI inside the chain of responsibility, not above it

In high-trust work, an AI experience must make three things clear: what evidence the system used, where a person must review it, and who remains responsible for the decision. The goal was a workflow useful enough to change the work and explicit enough to preserve human judgment.